CN118072492A - Electric power fire safety early warning method and comprehensive management platform - Google Patents

Abstract

The embodiment of the application provides an electric power fire safety early warning method and a comprehensive management platform, which are characterized in that through collecting electric power fire monitoring data of nodes in a key area, extracting key knowledge features and fusing the key knowledge features to generate an instant monitoring feature flow path, inputting the instant monitoring feature flow path into a fire risk pre-judging network, and finally generating a fire risk trend evolution graph, the accurate capturing and dynamic monitoring of the state change of a potential fire risk object in an electric power fire safety monitoring scene are realized, the fire risk condition can be found and early warned in time, and the monitoring efficiency and accuracy of the electric power fire safety are improved. Meanwhile, through generating multidimensional feature vector distribution and dynamic change vectors, the fire hazard trend can be adaptively identified and tracked in a complex and changeable electric fire-fighting environment, and powerful support is provided for fire-fighting safety management and emergency response.

Description

Electric power fire safety early warning method and comprehensive management platform

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to an electric power fire safety early warning method and a comprehensive management platform.

Background

With the rapid development of society and the continuous increase of power demand, the safe operation of power facilities has important significance for guaranteeing the life and property safety and social stability of people. However, due to potential safety hazards such as high temperature, smoke, harmful gas leakage and flame which may be generated during the operation of the electric power facilities, the electric power fire protection safety faces serious challenges. Therefore, it is important to develop a power fire safety early warning method capable of monitoring and early warning in real time and accurately judging the trend of fire.

Traditional electric power fire safety monitoring methods generally depend on manual inspection and simple sensor monitoring, and are not only low in efficiency, but also easily affected by human factors and environmental interference, so that the accuracy and timeliness of fire early warning cannot be guaranteed.

Disclosure of Invention

In order to at least overcome the defects in the prior art, the embodiment of the application aims to provide an electric fire safety early warning method and a comprehensive management platform.

According to one aspect of the present application, there is provided an electric fire safety precaution method, the method comprising:

Collecting power fire control monitoring data in a power fire control environment through a sensor network deployed at a key area node, wherein the power fire control monitoring data comprises at least one of temperature, smoke, gas concentration and flame brightness;

Extracting key knowledge features from the power fire control monitoring data according to a predefined power fire control safety monitoring knowledge base, wherein the key knowledge features are used for reflecting the state change of potential fire objects;

Fusing the extracted key knowledge features to generate multi-dimensional feature vector distribution, and generating an instant monitoring feature flow path of the electric fire safety monitoring scene according to dynamic change vectors in the multi-dimensional feature vector distribution;

Inputting an instant monitoring characteristic flow path of an electric power fire safety monitoring scene into a fire risk pre-judging network for completing knowledge learning, and generating a corresponding fire risk trend evolution graph, wherein the instant monitoring characteristic flow path is used for reflecting: the real-time characteristic change vector of each monitoring point in the electric power fire safety monitoring scene is used for reflecting the fire hazard trend evolution graph: and the trend evolution state characteristics of each potential fire hazard object related to the real-time characteristic change vector are obtained.

In a possible implementation manner of the first aspect, the training step of the fire risk prediction network includes:

According to the electric fire safety monitoring log sequence of the target fire safety tag, parameter learning is carried out on the initialized neural network, a first neural network for carrying out fire risk feature mining of the target fire safety tag is generated, parameter learning is carried out on the first neural network according to the first example monitoring feature flow path sequence, and a second neural network for carrying out real-time feature change vector extraction of the monitoring feature flow path is generated;

Performing parameter learning on the second neural network according to a second example monitoring characteristic flow path sequence to generate a third neural network for extracting potential fire risk objects and dangerous index features of the potential fire risk objects of the monitoring characteristic flow paths, wherein the second example monitoring characteristic flow path sequence comprises each second example monitoring characteristic flow path and corresponding potential fire risk object labeling data;

And performing parameter learning on the third neural network according to a third example monitoring characteristic flow path sequence to generate the fire risk pre-judging network, wherein the third example monitoring characteristic flow path sequence comprises all third example monitoring characteristic flow paths and corresponding marked fire risk trend evolution graphs.

In a possible implementation manner of the first aspect, the power fire safety monitoring log sequence includes a first power fire safety monitoring log sub-sequence and a second power fire safety monitoring log sub-sequence;

the parameter learning is performed on the initialized neural network according to the electric fire safety monitoring log sequence of the target fire safety tag, and a first neural network for performing fire risk feature mining of the target fire safety tag is generated, which comprises the following steps:

based on a non-guided knowledge learning strategy, carrying out parameter learning on the initialized neural network according to the first power fire safety monitoring log subsequence to generate a temporary neural network for carrying out fire risk embedding representation;

And according to the second electric fire safety monitoring log subsequence, parameter learning is carried out on the temporary neural network, a first neural network for carrying out fire risk feature mining is generated, and each electric fire safety monitoring log in the second electric fire safety monitoring log subsequence has corresponding fire risk feature labeling.

In a possible implementation manner of the first aspect, the non-guided knowledge learning strategy performs parameter learning on the initialized neural network according to the first power fire safety monitoring log subsequence, and generates a temporary neural network for performing fire risk embedding representation, which includes:

and according to the first power fire safety monitoring log subsequence, performing cyclic knowledge learning on the initialized neural network to generate the temporary neural network, wherein each round of cyclic knowledge learning performs the following steps:

Selecting a first electric fire safety monitoring log from the first electric fire safety monitoring log subsequence, carrying out initialization feature processing on the first electric fire safety monitoring log according to a preset non-guiding learning task, generating a processed first electric fire safety monitoring log, and constructing guiding annotation data corresponding to the first electric fire safety monitoring log;

Loading the processed first electric power fire safety monitoring log to the initialized neural network to generate corresponding fire safety embedded representation data;

generating a first error parameter according to the fire safety embedded representation data and the corresponding guiding annotation data, and performing knowledge learning on the initialized neural network according to the first error parameter.

In a possible implementation manner of the first aspect, the performing parameter learning on the temporary neural network according to the second power fire safety monitoring log sub-sequence to generate a first neural network for performing fire risk feature mining includes:

and according to the second power fire safety monitoring log subsequence, performing cyclic knowledge learning on the temporary neural network to generate the first neural network, wherein each round of cyclic knowledge learning performs the following steps:

Loading a second electric power fire safety monitoring log selected from the second electric power fire safety monitoring log subsequence to the temporary neural network to generate corresponding fire risk characteristics;

Generating a second error parameter according to the fire risk characteristic and the marked fire risk characteristic corresponding to the second electric power fire safety monitoring log, and performing knowledge learning on the temporary neural network according to the second error parameter.

In a possible implementation manner of the first aspect, the performing parameter learning on the first neural network according to the first exemplary monitoring feature flow path sequence, generating a second neural network for performing real-time feature variation vector extraction of the monitoring feature flow path, includes:

For each first example monitoring feature flow path in the sequence of first example monitoring feature flow paths, performing the following steps:

Obtaining each monitoring characteristic node information in a first example monitoring characteristic flow path and the flow state information of each monitoring characteristic node information, carrying out serialization coding on each monitoring characteristic node information according to the flow state information according to a preset coding rule to generate an example description section, and obtaining electric fire safety application environment information associated with the first example monitoring characteristic flow path;

According to the example description sections and the corresponding power fire safety application environment information, performing cyclic knowledge learning on the first neural network to generate the second neural network, wherein each round of cyclic knowledge learning performs the following steps:

Selecting an example description section from the example description sections, initializing the selected example description section according to a preset non-guiding learning task, generating a processed example description section, and constructing guiding annotation data corresponding to the example description section;

Loading the processed example description section and corresponding electric power fire safety application environment information to the first neural network to generate a corresponding monitoring characteristic flow path prediction result;

Generating a first error parameter according to the monitoring characteristic flow path prediction result and the corresponding guiding labeling data, and performing knowledge learning on the first neural network according to the first error parameter.

In a possible implementation manner of the first aspect, the performing parameter learning on the second neural network according to the second exemplary monitoring feature flow path sequence, to generate a third neural network for performing the monitoring feature flow path of the potential fire hazard object and the dangerous index feature extraction of the potential fire hazard object, includes:

Performing cyclic knowledge learning on the second neural network according to the second example monitoring characteristic flow path sequence to generate the third neural network, wherein each round of cyclic knowledge learning performs the following steps:

selecting a second example monitoring characteristic flow path from the second example monitoring characteristic flow path sequence, and obtaining a characteristic state change track of the second example monitoring characteristic flow path;

loading the characteristic state change track of the second example monitoring characteristic flow path to the second neural network to generate corresponding potential fire hazard object information, wherein the potential fire hazard object information comprises predicted potential fire hazard objects and dangerous index characteristics of the potential fire hazard objects;

And generating a third error parameter according to the potential fire hazard object information and the labeled potential fire hazard object data corresponding to the second example monitoring characteristic flow path, and performing knowledge learning on the second neural network according to the third error parameter.

In a possible implementation manner of the first aspect, the performing parameter learning on the third neural network according to the third exemplary monitoring feature flow path sequence to generate the fire risk prediction network includes:

Performing cyclic knowledge learning on the third neural network according to the third example monitoring characteristic flow path sequence to generate the fire risk pre-judging network, wherein each round of cyclic knowledge learning performs the following steps:

selecting a third example monitoring characteristic flow path from the third example monitoring characteristic flow path sequence, and obtaining a characteristic state change track of the third example monitoring characteristic flow path and potential fire hazard object information, wherein the potential fire hazard object information comprises all potential fire hazard objects and dangerous index characteristics of all the potential fire hazard objects related to the third example monitoring characteristic flow path;

loading the characteristic state change track of the third example monitoring characteristic flow path and the potential fire risk object information to the second neural network to generate a corresponding fire risk trend evolution graph;

Generating a fourth error parameter according to the fire risk trend evolution diagram and the marked fire risk trend evolution diagram corresponding to the third example monitoring characteristic flow path, and performing knowledge learning on the third neural network according to the fourth error parameter.

In a possible implementation manner of the first aspect, the inputting the real-time monitoring feature flow path of the electric power fire safety monitoring scene into the fire risk prediction network for completing knowledge learning, and generating a corresponding fire risk trend evolution graph includes:

Extracting each monitoring characteristic node information and corresponding flow state information in the instant monitoring characteristic flow path through a fire risk pre-judging network, and generating a characteristic state change track of the instant monitoring characteristic flow path according to each monitoring characteristic node information and corresponding flow state information;

and identifying each potential fire hazard object and the dangerous index characteristic of each potential fire hazard object related to the instant monitoring characteristic flow path by using the fire hazard pre-judging network according to the characteristic state change track, and generating a map according to the flow state information of the node information of each monitoring characteristic to generate a corresponding fire hazard trend evolution map.

For example, in a possible implementation manner of the first aspect, the power fire safety monitoring knowledge base defines various feature knowledge parameters related to fire risks, and combinations and weights of the feature parameters in different scenarios, and the step of extracting key knowledge features from the power fire safety monitoring data according to the predefined power fire safety monitoring knowledge base includes:

Selecting at least one characteristic knowledge parameter related to fire risk from the electric power fire safety monitoring knowledge base, and constructing a mapping relation between the electric power fire safety monitoring data and each selected characteristic knowledge parameter, wherein the mapping relation is a mathematical function, a logic rule or a machine learning model and is used for converting the electric power fire safety monitoring data into corresponding knowledge characteristic values;

Combining the mapping relation into a corresponding dynamic feature mapping matrix, inputting the power fire control monitoring data into the dynamic feature mapping matrix, and dynamically mapping to obtain corresponding initial key knowledge features;

Performing association analysis on the initial key knowledge features and entities and attributes in a preset expert knowledge graph, and determining potential contact data between different features in the initial key knowledge features;

and carrying out enhancement processing on the potential contact data to generate corresponding key knowledge features.

For example, in a possible implementation manner of the first aspect, the step of generating the real-time monitoring feature flow path of the electric fire safety monitoring scene according to the dynamic change vector in the multi-dimensional feature vector distribution includes:

mapping the multi-dimensional feature vector distribution into a high-dimensional feature space, wherein each dimension in the high-dimensional feature space represents a specific knowledge feature;

In the high-dimensional feature space, constructing a corresponding dynamic feature map according to a dynamic change relation between feature vectors in the multi-dimensional feature vectors, wherein the dynamic feature map reflects real-time association and evolution trend between different feature vectors;

Analyzing the change vector in the dynamic characteristic map, and identifying a key change vector with the influence weight on the electric power fire safety monitoring scene being greater than a preset weight, wherein the key change vector is used for representing a potential fire risk event or an abnormal situation;

Initializing a characteristic flow path for each key region node in the dynamic characteristic map, wherein the characteristic flow path consists of at least one key change vector and is used for reflecting the real-time state change of the key region node;

dynamically updating and tracking the characteristic flow path of each key area node according to whether the key change vector has a trigger monitoring condition larger than a preset threshold parameter value or not, wherein the method specifically comprises the steps of adding new characteristic points, deleting outdated characteristic points and adjusting the direction and speed of the path;

Fusing and optimizing the characteristic flow paths from different key area nodes to generate corresponding real-time monitoring characteristic flow path networks;

Displaying the real-time monitoring characteristic flow path to a monitoring person through a thermodynamic diagram, a flow diagram or an animation, and adjusting the display mode and the content of the real-time monitoring characteristic flow path in real time according to a feedback event between the monitoring person and the real-time monitoring characteristic flow path.

According to one aspect of the embodiments of the present application, there is provided an integrated management platform comprising a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the power fire safety precaution method of any one of the foregoing possible implementations.

According to an aspect of embodiments of the present application, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions are read from a computer-readable storage medium by a processor of a computer device, which executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations of the three aspects described above.

According to the technical scheme provided by the embodiments of the application, the embodiment of the application extracts key knowledge features and fuses the key knowledge features by collecting the power fire control monitoring data of the key area nodes to generate the instant monitoring feature flow path, and then inputs the instant monitoring feature flow path into the fire risk prejudging network to finally generate the fire risk trend evolution graph, so that the accurate capturing and dynamic monitoring of the state change of the potential fire risk object in the power fire control safety monitoring scene are realized, the fire risk condition can be found and early warned in time, and the monitoring efficiency and accuracy of the power fire control safety are improved. Meanwhile, through generating multidimensional feature vector distribution and dynamic change vectors, the fire hazard trend can be adaptively identified and tracked in a complex and changeable electric fire-fighting environment, and powerful support is provided for fire-fighting safety management and emergency response.

Drawings

For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated in the embodiments, and it should be understood that the following drawings only illustrate some embodiments of the present application, and therefore should not be considered as limiting the scope, and other related drawings can be extracted by those skilled in the art without the inventive effort.

FIG. 1 is a schematic flow chart of a method for early warning of electric fire safety according to an embodiment of the present application;

Fig. 2 is a schematic block diagram of an architecture of a comprehensive management platform for implementing the above-mentioned electric fire safety pre-warning method according to an embodiment of the present application.

Detailed Description

The following description is presented to enable one of ordinary skill in the art to make and use the application and is provided in the context of a particular application and its requirements. It will be apparent to those having ordinary skill in the art that various changes can be made to the disclosed embodiments and that the general principles defined herein may be applied to other embodiments and applications without departing from the principles and scope of the application. Therefore, the present application is not limited to the described embodiments, but is to be accorded the widest scope consistent with the claims.

Fig. 1 is a schematic flow chart of an electric fire safety pre-warning method according to an embodiment of the present application, and the electric fire safety pre-warning method is described in detail below.

Step S110, collecting power fire control monitoring data in a power fire control environment through a sensor network deployed at a key area node, wherein the power fire control monitoring data comprises at least one of temperature, smoke, gas concentration and flame brightness.

In detail, the critical area node refers to a geographical location or a point of equipment in an electric power facility that is critical to fire safety. These critical area nodes are typically areas where the risk of fire is high or the consequences of a fire are severe once they occur. For example, in power substations, transformer rooms, switchrooms, cable trenches, etc. are critical area nodes where deployment of sensor networks can more effectively monitor potential fire risks.

The sensor network is a network composed of a plurality of sensor nodes, and the nodes are distributed in a key area and can monitor and collect environmental parameters in real time. For example, a temperature sensor network may include tens or even hundreds of temperature sensors distributed throughout corners of an electrical utility that transmit real-time temperature data to a central integrated management platform via a wireless network.

The power fire control monitoring data refer to various environmental parameter data which are collected through a sensor network and are related to the fire safety of the power facilities. For example, the power fire monitoring data may include information such as real-time temperature of the transformer room, smoke concentration in the cable tunnel, concentration of harmful gases (e.g., carbon monoxide) in the distribution room, and flame intensity captured by the infrared camera. Specifically, the temperature refers to the heat of the environment, typically expressed in degrees celsius or degrees fahrenheit. The smoke refers to tiny particles or aerosols suspended in air, typically generated by combustion. The gas concentration refers to the content or proportion of a specific gas in the air, such as oxygen, carbon monoxide, etc. The flame brightness refers to the intensity of visible light emitted by the flame, and is generally related to the temperature of the flame and the burning substance. For example, in an electrical installation, if a temperature sensor detects abnormally high temperatures, a smoke detector detects an increase in smoke concentration, a gas analyzer detects an increase in harmful gas concentration, or a camera captures a sudden increase in flame brightness, which may be an impending sign of a fire.

Thus, in this embodiment, in a large power facility, sensor networks are deployed at key regional nodes such as transformer rooms, cable tunnels, distribution rooms, and the like. The sensor networks can monitor and collect power fire control monitoring data such as temperature, smoke concentration, harmful gas concentration (such as carbon monoxide, hydrogen sulfide and the like) and flame brightness and the like in the environment in real time. The integrated management platform can be connected with the sensor networks in a wired or wireless mode, and receives the power fire control monitoring data at regular time or in real time.

And step S120, extracting key knowledge features from the power fire control monitoring data according to a predefined power fire control safety monitoring knowledge base, wherein the key knowledge features are used for reflecting the state change of potential fire objects.

In detail, the power fire safety monitoring knowledge base is a predefined and constructed knowledge set, and comprises various rules, standards, experiences, models and the like related to the fire safety of the power facilities. For example, the power fire safety monitoring knowledge may include information about a standard of overheating of a cable, a method for identifying leakage of transformer oil, a prediction model of a fire spread speed, and the like. This information is critical to extracting key knowledge features from the monitored data.

The key knowledge features are important information or features extracted from the power fire control monitoring data and can directly reflect the state change of the potential fire hazard objects. For example, if the temperature sensor data shows a continuously rapid rise in temperature for an area, this "continuously rapid rise" temperature change is a key knowledge feature that may indicate that the area is at risk of a potential fire. Also, sudden increases in smoke concentration, abnormal increases in harmful gas concentration, or abrupt increases in flame brightness may all be key knowledge features.

Thus, in this embodiment, the integrated management platform uses these power fire safety monitoring knowledge bases to process and analyze the power fire monitoring data collected from the sensor network. For example, key knowledge features such as abnormal rising trend, abrupt change point of smoke concentration, rapid increase of harmful gas concentration, appearance of flame brightness and the like in the temperature data can be extracted. These key knowledge features can reflect changes in the state of potential fire hazard objects (e.g., cables, transformer oil, insulation, etc.).

And step S130, fusing the extracted key knowledge features to generate multi-dimensional feature vector distribution, and generating an instant monitoring feature flow path of the electric fire safety monitoring scene according to dynamic change vectors in the multi-dimensional feature vector distribution.

In detail, the multidimensional feature vector distribution refers to a high-dimensional data representation formed by fusing a plurality of key knowledge features. In this multi-dimensional feature vector distribution, each dimension represents a particular knowledge feature, and the entire multi-dimensional feature vector distribution reflects the combined performance of these features in the power fire safety monitoring scenario. For example, assuming that three key knowledge features of temperature, smoke concentration, and flame brightness are extracted, the values of these three features may be combined into a three-dimensional feature vector (e.g., [ temperature value, smoke concentration value, flame brightness value ]). Over time, these feature vectors form a multi-dimensional feature vector distribution that reflects the variation of various parameters in the monitored scene.

The motion change vector refers to a movement or change of the feature vector that occurs with time in the multidimensional feature vector distribution. These changes may reflect the status trend of potential fire objects in the power fire safety monitoring scenario. For example, if the value of the temperature characteristic continues to rise over a period of time, and the smoke concentration and flame intensity also exhibit a trend toward an increase, these changes may be represented as a dynamic change vector, pointing in the direction of "temperature rise, smoke concentration increase, flame intensity increase". The dynamic change vector can help judge the development trend of the fire risk.

The real-time monitoring characteristic flow path is a data flow path generated according to real-time characteristic change vectors of monitoring points in the electric power fire safety monitoring scene. The data flow path can reflect the state change of each monitoring point in the monitoring scene in real time, and provides a basis for the instant pre-judgment of fire risks. For example, assume that a network of sensors is deployed in multiple critical areas of an electrical facility, each of which will collect and send data to an integrated management platform in real time. The comprehensive management platform generates an instant characteristic flow path according to the data, and the characteristic flow path contains the change conditions of the characteristics such as the temperature, the smoke concentration and the like of each monitoring point. By observing this path, the potential fire risk can be found in time and corresponding measures can be taken.

Therefore, in this embodiment, the integrated management platform fuses the extracted key knowledge features to form a multidimensional feature vector distribution. This multi-dimensional feature vector distribution not only contains readings of a single sensor, but also fuses correlation information and time series data between multiple sensors. Therefore, the comprehensive management platform further analyzes dynamic change vectors in the multidimensional feature vector distribution, such as continuous temperature rise, smoke concentration diffusion trend and the like, and generates an instant monitoring feature flow path of the electric power fire safety monitoring scene. The real-time monitoring characteristic flow path can reflect characteristic change vectors of all monitoring points in real time, and helps the comprehensive management platform to judge fire hazard conditions more accurately.

Step S140, inputting an instant monitoring characteristic flow path of the electric power fire safety monitoring scene into a fire risk pre-judging network for completing knowledge learning, and generating a corresponding fire risk trend evolution graph, wherein the instant monitoring characteristic flow path is used for reflecting: the real-time characteristic change vector of each monitoring point in the electric power fire safety monitoring scene is used for reflecting the fire hazard trend evolution graph: and the trend evolution state characteristics of each potential fire hazard object related to the real-time characteristic change vector are obtained.

In detail, the fire risk prediction network is a network model constructed by using a machine learning algorithm, and can predict the development trend of fire risk according to the input instant monitoring characteristic flow path. The fire risk pre-judging network can identify modes and features related to fire risks by learning and analyzing a large amount of historical data, and accordingly pre-judge the risks. For example, a fire risk prediction network based on deep learning can predict the probability of fire occurrence in a future period of time according to the input characteristic data such as temperature, smoke concentration and the like. When the comprehensive management platform receives new real-time data, the data are input into a fire risk pre-judging network, and the fire risk pre-judging network outputs a trend evolution graph of fire risks, so that the state change trend of potential fire risk objects can be known, and preventive measures can be taken in time to avoid fire occurrence.

That is, in this embodiment, the integrated management platform may input the real-time monitoring feature flow path into a fire risk prediction network that has completed knowledge learning. The fire risk pre-judging network processes and analyzes the input immediate monitoring characteristic flow path through algorithms such as deep learning and the like to generate a corresponding fire risk trend evolution graph. The fire risk trend evolution graph not only shows the real-time characteristic change vectors of all monitoring points, but also predicts the trend evolution state characteristics of all potential fire risk objects related to the real-time characteristic change vectors. For example, it may be predicted that the temperature of an area will continue to rise, smoke will gradually spread, and fire levels will continue to rise. According to the fire hazard trend evolution diagram, the comprehensive management platform can discover fire hazard hidden dangers in time and take corresponding preventive measures.

Based on the steps, the embodiment of the application extracts key knowledge features by collecting the power fire control monitoring data of the key area nodes, fuses the key knowledge features, generates an instant monitoring feature flow path, inputs the instant monitoring feature flow path into the fire risk prediction network, and finally generates a fire risk trend evolution graph, thereby realizing the accurate capturing and dynamic monitoring of the state change of the potential fire risk object in the power fire control safety monitoring scene, timely finding and early warning the fire risk situation, and improving the monitoring efficiency and accuracy of the power fire control safety. Meanwhile, through generating multidimensional feature vector distribution and dynamic change vectors, the fire hazard trend can be adaptively identified and tracked in a complex and changeable electric fire-fighting environment, and powerful support is provided for fire-fighting safety management and emergency response.

For example, in one possible implementation, the power fire safety monitoring knowledge base defines various characteristic knowledge parameters related to fire risk, and combinations and weights of the characteristic parameters in different scenarios, and step S120 may include:

Step S121, selecting at least one feature knowledge parameter related to fire risk from the power fire protection security monitoring knowledge base, and constructing a mapping relationship with the power fire protection monitoring data for each selected feature knowledge parameter, where the mapping relationship is a mathematical function, a logic rule or a machine learning model, and is used to convert the power fire protection monitoring data into corresponding knowledge feature values.

In this embodiment, the integrated management platform first accesses a power fire safety monitoring knowledge base, which has predefined various characteristic knowledge parameters related to fire risk, such as a temperature threshold, a smoke concentration increase rate, the presence of a specific gas, and the like. These characteristic knowledge parameters are derived based on historical data, expert experience and industry standards and are critical for judging fire risk.

The comprehensive management platform selects a group of characteristic knowledge parameters which are most relevant to the current monitoring scene from the electric power fire safety monitoring knowledge base, for example, selects temperature, smoke concentration and flame brightness as key monitoring indexes. Then, the integrated management platform constructs a mapping relation for each selected characteristic knowledge parameter. For temperature, the mapping may be a mathematical function that converts degrees celsius to kelvin; for smoke concentration, there may be a machine learning model trained based on historical data for converting raw sensor readings into normalized smoke concentration values; for flame brightness, there may be a simple threshold logic rule that beyond a certain brightness value, a flame is considered to be present.

Step S122, combining the mapping relations into corresponding dynamic feature mapping matrixes, inputting the power fire control monitoring data into the dynamic feature mapping matrixes, and dynamically mapping to obtain corresponding initial key knowledge features.

In this embodiment, the integrated management platform combines the constructed mapping relationships into a dynamic feature mapping matrix. The dynamic feature mapping matrix can dynamically map out corresponding initial key knowledge features according to the input power fire control monitoring data. For example, when the integrated management platform receives a new temperature reading, it will use the temperature mapping relationship in the dynamic feature mapping matrix to convert this reading to a kelvin temperature value; similarly, when data of smoke concentration and flame brightness are received, the corresponding mapping relation is also used for conversion.

In this way, the integrated management platform can convert the original power fire control monitoring data into initial key knowledge features with more meaning and comparability in real time.

And step S123, performing association analysis on the initial key knowledge features and the entities and attributes in the preset expert knowledge patterns, and determining potential contact data between different features in the initial key knowledge features.

In this embodiment, the integrated management platform performs the association analysis on the initial key knowledge features and the preset expert knowledge patterns. The preset expert knowledge graph is a graph-shaped data structure containing various entities (such as cables, transformers and the like) and attributes (such as material types, insulation grades and the like) of the electric power facilities. The map is constructed based on knowledge and experience of field experts and is used for assisting the comprehensive management platform in understanding the relationship and the attribute among the entities in the monitoring scene.

Through carrying out association analysis on the initial key knowledge features and the entities and attributes in the preset expert knowledge patterns, the comprehensive management platform can find potential relations among different features. For example, the integrated management platform may find that an abnormal rise in temperature in a region is related to the extent of aging of the regional cable; or an increase in smoke concentration may be related to oil leakage from nearby transformers. These potential links provide a deeper view of fire risk analysis for the integrated management platform.

And step S124, carrying out enhancement processing on the potential contact data to generate corresponding key knowledge features.

And finally, the comprehensive management platform performs enhancement processing on the potential contact data to generate final key knowledge features. The enhancement processing may include steps of data cleaning, feature selection, feature fusion, etc., in order to improve accuracy and robustness of key knowledge features.

For example, the integrated management platform may remove abnormal data generated due to sensor faults or false alarms; or by combining multiple related potential contact data into a more representative feature through feature fusion. After enhancement processing, the comprehensive management platform obtains a group of high-quality key knowledge features which can accurately reflect the state change of each potential fire hazard object in the electric power fire safety monitoring scene.

Based on the steps, the embodiment realizes the efficient extraction of key knowledge features in the power fire control monitoring data by constructing the power fire control safety monitoring knowledge base, defining the feature knowledge parameters related to the fire risk and the combination and weight of the feature knowledge parameters under different scenes. By constructing a mapping relation between each selected characteristic knowledge parameter and the power fire control monitoring data, the original monitoring data can be converted into knowledge characteristic values with higher expressive force. Further, the dynamic feature mapping matrix is formed by combining the mapping relations, so that initial key knowledge features can be dynamically mapped and obtained, and the requirements of different monitoring scenes are met. In addition, through carrying out association analysis with the entity and the attribute in the preset expert knowledge graph, potential contact data among different features in the initial key knowledge features can be revealed, and further the potential contact data is enhanced, so that more accurate and comprehensive key knowledge features are generated. Therefore, the accuracy and the reliability of fire risk early warning are improved.

For example, in one possible implementation, step S130 may include:

Step S131, mapping the multi-dimensional feature vector distribution into a high-dimensional feature space, where each dimension in the high-dimensional feature space represents a specific knowledge feature.

In this embodiment, in this high-dimensional feature space, the relationship between feature vectors, which are otherwise difficult to be linearly distinguished, becomes clearer in the multi-dimensional feature vector distribution. For example, by mapping, the integrated management platform may find that as temperature and smoke concentration rise simultaneously, their distances in the high-dimensional feature space will be closer, indicating a strong correlation between them.

And S132, constructing a corresponding dynamic characteristic map according to the dynamic change relation between the characteristic vectors in the multi-dimensional characteristic vectors in the high-dimensional characteristic space, wherein the dynamic characteristic map reflects the real-time association and evolution trend between different characteristic vectors.

In the high-dimensional feature space, the integrated management platform starts to construct a dynamic feature map. The dynamic feature map records not only the position of each feature vector at a certain moment, but also the change track of each feature vector along with time. To achieve this, the integrated management platform may use a Dynamic Time Warping (DTW) algorithm to calculate the similarity and distance between different feature vectors.

Over time, this dynamic profile will continually update and evolve. For example, as the temperature of a region continues to rise, the feature vector associated therewith may form a distinct hot spot in the dynamic profile, indicating that the region may be at risk for a potential fire.

And step S133, analyzing the change vector in the dynamic characteristic map, and identifying a key change vector with the influence weight on the electric power fire safety monitoring scene being greater than the preset weight, wherein the key change vector is used for representing a potential fire hazard event or abnormal situation.

In this embodiment, the comprehensive management platform performs deep analysis on the change vectors in the dynamic feature map, firstly calculates the change speed and acceleration of each feature vector, and then screens out the key change vectors with the greatest influence on the electric fire safety according to the preset weight and threshold.

For example, the integrated management platform finds that the temperature in a certain cable well rises sharply in a short time, accompanied by a rapid increase in smoke concentration. Both the change speed and the acceleration of the two feature vectors exceed a preset threshold value and are thus identified as key change vectors. These key change vectors indicate that there may be serious fire hazards within the cable pit.

Step S134, initializing a feature flow path for each key region node in the dynamic feature map, where the feature flow path is composed of at least one key change vector and is used to reflect the real-time state change of the key region node.

After the key change vectors are identified, the integrated management platform begins initializing a feature flow path for each key region node. This characteristic flow path is in fact a sequence of key change vectors describing the evolution of the region node from the initial state to the current state.

Taking a cable well as an example, the integrated management platform firstly determines the initial state (such as temperature, smoke concentration and the like) of the cable well, and then gradually updates the state according to the key change vector monitored in real time. These status updates are recorded and form a continuous characteristic flow path reflecting the progression of fire hazards in the cable well.

Step S135, dynamically updating and tracking the characteristic flow path of each key area node according to whether the key change vector has trigger monitoring conditions larger than a preset threshold parameter value, and specifically comprises adding new characteristic points, deleting outdated characteristic points and adjusting the direction and speed of the path.

Once the feature flow paths are initialized, the integrated management platform will begin to dynamically update and track them. For example, it may involve monitoring new key change vectors in real-time and adding them to existing feature flow paths, and periodically evaluating the validity and accuracy of existing paths.

For example, as the temperature within the cable well continues to rise and trigger a new alarm, the integrated management platform will immediately add this new key change vector to the characteristic flow path of the zone. Meanwhile, the direction and the speed of the path can be adjusted according to the latest monitoring data so as to more accurately reflect the actual development condition of the fire hidden danger.

And step S136, fusing and optimizing the characteristic flow paths from the nodes in different key areas to generate a corresponding real-time monitoring characteristic flow path network.

Over time, more and more feature flow paths are created and updated. To more effectively manage and utilize these characteristic flow paths, the integrated management platform begins to fuse the characteristic flow paths into a unified, instant monitoring characteristic flow path network. In this process, the integrated management platform may use mathematical tools of graph theory to describe the relationships and connections between the different feature flow paths.

The comprehensive management platform can more comprehensively know the fire safety condition of the whole electric power facility through fusing and optimizing the characteristic flow path network, not only can the fire hidden danger development condition of each key area node be monitored in real time, but also the fire danger event possibly happening in the future can be predicted, and corresponding precautionary measures can be adopted in advance.

Step S137, displaying the real-time monitoring feature flow path to a monitoring person through a thermodynamic diagram, a flow diagram or an animation, and adjusting the displaying mode and the content of the real-time monitoring feature flow path in real time according to a feedback event between the monitoring person and the real-time monitoring feature flow path.

And finally, the integrated management platform displays the real-time monitoring characteristic flow path to monitoring personnel in a visual mode. These visualization tools include thermodynamic diagrams, flow charts, animations, etc., which can help monitoring personnel to intuitively understand the fire safety conditions of the electrical utility and potential fire events.

Meanwhile, the comprehensive management platform also provides rich interaction functions, and allows monitoring personnel to adjust the display mode and content of the characteristic flow path according to actual demands. For example, they may better focus on and analyze potential fire events by zooming in on a certain area, adjusting color schemes, or adding notes. In addition, the comprehensive management platform can adjust the display mode and the content of the characteristic flow path in real time according to the feedback of monitoring personnel so as to meet different monitoring requirements and improve the working efficiency.

In one possible implementation, step S140 may include:

Step S141, extracting each monitoring characteristic node information and corresponding flow state information in the instant monitoring characteristic flow path through a fire risk pre-judging network, and generating a characteristic state change track of the instant monitoring characteristic flow path according to each monitoring characteristic node information and corresponding flow state information.

In this embodiment, the integrated management platform first inputs the real-time monitoring characteristic flow path of the electric power fire safety monitoring scene into the fire risk prediction network that has completed knowledge learning. The fire risk pre-judging network is obtained through a large amount of historical data and expert knowledge training, and has the capability of accurately pre-judging the fire risk.

And extracting the information of each monitoring characteristic node in the real-time monitoring characteristic flow path by the comprehensive management platform by using a deep learning algorithm in the fire risk pre-judging network, wherein the information of each monitoring characteristic node comprises real-time data of key physical quantities such as temperature, smoke concentration, gas components and the like. Meanwhile, the integrated management platform also extracts flow state information among the characteristic nodes, such as a change trend, a speed and the like of data. The information is integrated to form a characteristic state change track of the real-time monitoring characteristic flow path. The track not only records the real-time state of each characteristic node, but also reflects the dynamic relationship and evolution trend among the characteristic nodes.

Step S142, identifying each potential fire object and the dangerous index feature of each potential fire object related to the real-time monitoring feature flow path according to the feature state change track by using the fire risk pre-judging network, and generating a map according to the flow state information of the node information of each monitoring feature to generate a corresponding fire trend evolution map.

And then, the comprehensive management platform utilizes a fire risk prejudging network to conduct deep analysis on the characteristic state change track. By comparing the historical data with a preset fire risk model, the integrated management platform can identify each potential fire risk object related to the instant monitoring feature flow path.

These potential fire hazard objects may be a particular device, area, or system, such as a transformer, cable pit, distribution room, etc. For each potential fire hazard object, the comprehensive management platform further identifies dangerous index features such as abnormal temperature rise, excessive smoke concentration and the like.

After identifying the potential fire hazard objects and the dangerous index features, the integrated management platform starts to display the information in a graphical mode. Firstly, according to the flow state information of each monitoring characteristic node, determining the basic structure and layout of the fire hazard trend evolution graph.

The integrated management platform then adds the individual potential fire objects and their hazard index features in the form of nodes and edges to the fire trend evolution graph. The size, color and other attributes of the nodes can be dynamically adjusted according to the severity of the dangerous index features, so that the actual situation of fire danger can be reflected more intuitively. And finally, the comprehensive management platform generates a final fire trend evolution graph by using a graph rendering technology and sends the final fire trend evolution graph to monitoring personnel for checking and analyzing. Through the fire hazard trend evolution diagram, monitoring personnel can clearly see the positions, the dangerous degrees and the association relations among all potential fire hazard objects, so that corresponding precaution measures can be taken in time to reduce the fire hazard.

Based on the steps, the embodiment realizes the accurate identification and real-time tracking of potential fire events or abnormal conditions in the electric power fire safety monitoring scene by mapping the multidimensional feature vector distribution in the high-dimensional feature space and constructing the dynamic feature map according to the dynamic change relation among the feature vectors. By analyzing the change vector in the dynamic characteristic map, the key change vector with great weight on the monitored scene can be identified, so that potential fire danger events can be effectively represented. Further, initializing a characteristic flow path for each key area node, dynamically updating and tracking the characteristic flow path according to the triggering monitoring condition of the key change vector, and ensuring that the state change of the key area node is reflected in real time. By fusing and optimizing the characteristic flow paths from the nodes of different key areas, an instant monitoring characteristic flow path network is generated, and an intuitive and comprehensive fire hazard monitoring view is provided for monitoring personnel. In addition, the invention can also adjust the display mode and the content of the characteristic flow path in real time according to the feedback of the monitoring personnel, thereby improving the monitoring efficiency and the user experience. Therefore, the method has remarkable effect in improving fire risk early warning accuracy and instantaneity.

In one possible implementation, the training step of the fire risk prediction network includes:

Step A110, performing parameter learning on an initialized neural network according to an electric fire safety monitoring log sequence of a target fire safety tag, generating a first neural network for performing fire risk feature mining of the target fire safety tag, performing parameter learning on the first neural network according to a first example monitoring feature flow path sequence, and generating a second neural network for performing real-time feature change vector extraction of a monitoring feature flow path.

Illustratively, at an electrical utility monitoring center, the integrated management platform receives and stores a plurality of sequences of electrical fire safety monitoring logs. These sequences of electrical fire safety monitoring logs detail critical fire safety related data such as temperature, smoke concentration, gas composition, etc. of the electrical facility, and each electrical fire safety monitoring log is accompanied by one or more target fire safety labels such as "normal", "high temperature warning", "smoke over standard", etc.

The comprehensive management platform uses the electric power fire safety monitoring log sequences with the labels as a training data set, initializes a neural network model and starts parameter learning. In the training process, the comprehensive management platform continuously adjusts the weight and bias of the network by comparing the difference between the network output and the actual label, so that the network can gradually learn to extract the characteristics related to fire safety from the log data and accurately classify the characteristics.

After multiple rounds of iterative training, the integrated management platform generates a first neural network model, namely a first neural network. The first neural network can automatically mine the characteristics related to the fire risk according to the input electric power fire safety monitoring log sequence, and provides basic data support for subsequent fire risk pre-judgment.

To further enhance the performance of the first neural network, the integrated management platform obtains a set of first example monitoring feature flow path sequences. These first example monitoring signature flow path sequences are extracted from actual power utility monitoring scenarios, each sequence containing a series of consecutive signature nodes and flow state information between them, such as trend of temperature change, smoke diffusion path, etc.

The integrated management platform inputs the first example monitoring feature flow path sequences into the first neural network for further parameter learning. In the training process, the comprehensive management platform mainly focuses on the extraction capability of the network to the real-time feature change vector in the feature flow path. By comparing the difference between the network output and the actual characteristic change vector, the comprehensive management platform continuously adjusts the parameters of the network, so that the network can more accurately capture the dynamic change information in the characteristic flow path.

After this round of training, the first neural network is optimized to be a second neural network, which not only can extract fire risk features from the log data, but also can effectively extract and analyze real-time feature change vectors in the feature flow path.

Step a120, performing parameter learning on the second neural network according to a second example monitoring feature flow path sequence, and generating a third neural network for performing potential fire hazard objects of the monitoring feature flow paths and dangerous index feature extraction of the potential fire hazard objects, where the second example monitoring feature flow path sequence includes each second example monitoring feature flow path and corresponding labeled potential fire hazard object data.

In this embodiment, to further enhance the ability of the neural network to identify potential fire hazard objects and hazard index features, the integrated management platform obtains a set of second example monitoring feature flow path sequences. Unlike the first example monitored feature flow path sequence, the second example monitored feature flow path sequence also includes labeled potential fire hazard object data that indicates which feature nodes or regions are potentially at fire hazard and the severity of the hazard.

The integrated management platform inputs the second example monitoring feature flow path sequences into a second neural network for parameter learning. In the training process, the comprehensive management platform continuously adjusts parameters of the network by comparing the difference between the network output and the actual marking data, so that the network can more accurately identify potential fire hazard objects and dangerous index features thereof from the characteristic flow path.

After this round of training, the second neural network is optimized to be a third neural network, which not only can extract the real-time feature change vector, but also can accurately identify the potential fire hazard objects and corresponding dangerous index features in the monitoring feature flow path.

And step A130, performing parameter learning on the third neural network according to a third example monitoring characteristic flow path sequence to generate the fire risk prediction network, wherein the third example monitoring characteristic flow path sequence comprises each third example monitoring characteristic flow path and a corresponding marked fire risk trend evolution graph.

Finally, in order to enable the neural network to directly output the fire trend evolution graph, the integrated management platform obtains a set of third example monitoring feature flow path sequences. The third example monitoring feature flow path sequence contains marked fire risk trend evolution graph data, and the development trend and the evolution process of the fire risk are intuitively displayed.

The integrated management platform inputs the third example monitoring characteristic flow path sequences into a third neural network for final parameter learning. In the training process, the comprehensive management platform continuously adjusts parameters and structures of the network by comparing the difference between the fire risk trend evolution graph output by the network and the actual annotation graph, so that the network can accurately generate a corresponding fire risk trend evolution graph according to the input characteristic flow path data.

After this round of training, the third neural network is finally optimized as a fire risk prediction network. The network not only has strong feature extraction and classification capability, but also can directly output a visual fire hazard trend evolution diagram, and provides comprehensive and effective technical support for electric power fire safety monitoring. In practical application, the integrated management platform can utilize the third neural network to process and analyze the real-time electric power fire safety monitoring data, discover potential fire risks in time and take corresponding precautionary measures.

Based on the steps, the embodiment realizes the accurate identification and trend prejudgment of potential fire hazard objects in the electric power fire safety monitoring scene by constructing and training the fire hazard risk prejudging network. Specifically, a first neural network is trained according to a power fire safety monitoring log sequence of a target fire safety tag and used for mining fire risk features related to the target fire safety tag. Further, parameter learning is performed on the first neural network through the first example monitoring feature flow path sequence, so that a second neural network is obtained, and the second neural network can extract feature change vectors in the monitoring feature flow paths in real time. The second neural network is then further learned using the second example monitoring feature flow path sequence to generate a third neural network that is capable of extracting potential fire hazard objects and their hazard index features in the monitoring feature flow path. Finally, training a third neural network through a third example monitoring characteristic flow path sequence to obtain a fire risk pre-judging network, wherein the fire risk pre-judging network can accurately generate output reflecting a fire trend evolution graph. Therefore, the fire risk pre-judging network has remarkable effect in improving the accuracy and the instantaneity of the electric fire safety pre-warning.

In one possible embodiment, the power fire safety monitoring log sequence includes a first power fire safety monitoring log sub-sequence and a second power fire safety monitoring log sub-sequence.

Step a110 may include:

and step A111, based on a non-guided knowledge learning strategy, carrying out parameter learning on the initialized neural network according to the first power fire safety monitoring log subsequence, and generating a temporary neural network for carrying out fire risk embedding representation.

Illustratively, at a large power facility monitoring center, the integrated management platform continually receives and processes fire safety monitoring log data from various monitoring points. These fire safety monitoring log data are divided into two categories: the first power fire safety monitoring log subsequence and the second power fire safety monitoring log subsequence. The first power fire safety monitoring log sub-sequence mainly comprises daily monitoring data without specific labels, and the second power fire safety monitoring log sub-sequence comprises log data which are marked by experts and have specific fire risk characteristics.

The integrated management platform firstly adopts a non-guiding knowledge learning strategy, which means that the integrated management platform trains and initializes the neural network by using the first power fire safety monitoring log subsequence without labels, so that the integrated management platform can learn the basic structure and mode in log data. This learning strategy does not rely on a priori knowledge or labels, but rather lets the initializing neural network explore the intrinsic rules in the data itself.

The comprehensive management platform inputs the data in the first power fire safety monitoring log subsequence into the initialized neural network in batches, and parameters such as weight and bias of the network are continuously adjusted through forward propagation and backward propagation algorithms, so that the network can gradually learn basic feature representation in the log data. This process may require multiple iterations until the network reaches a certain steady state.

Through the training of the step, the integrated management platform generates a temporary neural network which has the capability of extracting basic characteristics from the original log data and carrying out embedded representation.

And step A112, performing parameter learning on the temporary neural network according to the second electric fire safety monitoring log subsequence, and generating a first neural network for performing fire risk feature mining, wherein each electric fire safety monitoring log in the second electric fire safety monitoring log subsequence has a corresponding fire risk feature labeling.

Next, the integrated management platform will utilize the second power fire safety monitoring log subsequence with the fire risk signature to further learn parameters of the temporary neural network. Unlike the first step, this step is supervised learning, since the integrated management platform already knows the fire risk profile labels for each log sample.

The comprehensive management platform inputs the data in the second electric power fire safety monitoring log subsequence and the corresponding label into the temporary neural network, and adjusts the parameters of the temporary neural network by comparing the difference between the network output and the actual label (such as using a cross entropy loss function). In this way, the temporary neural network is able to learn not only the basic feature representations of the log data, but also to learn to map these features onto specific fire risk tags.

After multiple rounds of iterative training, the temporary neural network is optimized as a first neural network. The first neural network has strong fire risk feature mining capability, can automatically extract key features related to fire risk according to input electric power fire safety monitoring log data, and provides powerful support for subsequent fire risk pre-judgment.

In one possible embodiment, step a111 may include:

step A1111, performing cyclic knowledge learning on the initialized neural network according to the first power fire safety monitoring log subsequence, and generating the temporary neural network, wherein each round of cyclic knowledge learning performs the following steps:

Step A1112, selecting a first electric fire safety monitoring log from the first electric fire safety monitoring log subsequence, performing initialization feature processing on the first electric fire safety monitoring log according to a preset non-guidance learning task, generating a processed first electric fire safety monitoring log, and constructing guidance marking data corresponding to the first electric fire safety monitoring log.

And step A1113, loading the processed first electric power fire safety monitoring log to the initialized neural network to generate corresponding fire safety embedded representation data.

And step A1114, generating a first error parameter according to the fire safety embedded representation data and the corresponding guiding annotation data, and performing knowledge learning on the initialized neural network according to the first error parameter.

For example, in order to construct a neural network model capable of automatically identifying and predicting fire risks, the integrated management platform first adopts a non-guided knowledge learning strategy, and the embodiment can train by using a first power fire safety monitoring log sub-sequence without a tag.

Firstly, the comprehensive management platform randomly selects a log record from a first power fire safety monitoring log subsequence, wherein the log record contains fire safety related data such as temperature, smoke concentration, gas composition and the like of a certain monitoring point in a certain time period. These data are raw, and require further feature extraction and conversion to be used by the neural network.

And then, the comprehensive management platform performs initialization feature processing on the selected first electric power fire safety monitoring log according to a preset non-guided learning task. This task may include data cleaning, feature extraction and conversion steps. For example, the integrated management platform may remove redundant information in the first power fire safety monitoring log, populate missing values, and convert the raw data into a numerical format that can be processed by the neural network.

In the process, the integrated management platform also builds the guiding annotation data corresponding to the first electric power fire safety monitoring log. Although these first power fire safety monitoring logs are not labeled per se, the integrated management platform can generate some preliminary labels or labeling data by some unsupervised learning methods (such as clustering algorithm) as a reference for the subsequent training process.

After the feature processing is completed, the integrated management platform loads the processed first power fire safety monitoring log into the initialized neural network. The initialized neural network is a deep learning model, has a multi-layer neuron structure, and can automatically extract and represent complex characteristics of data.

By means of the forward propagation algorithm, the neural network is initialized to convert the input log data into a form called "embedded representation". This embedded representation is an internal representation of the input data by the neural network, which captures the underlying structure and pattern of the data, providing useful characterization information for subsequent tasks.

The comprehensive management platform compares the generated fire safety embedded representation data with the guiding annotation data constructed before, and calculates the difference or error between the fire safety embedded representation data and the guiding annotation data. This error reflects the gap between the current understanding of the input data by the neural network and the actual situation.

To reduce this error, the integrated management platform uses a back-propagation algorithm and gradient descent optimizer to adjust parameters (e.g., weights and biases) of the neural network. Through a number of iterations of this process, the neural network gradually learns how to better represent and understand the fire safety features in the input data.

After a certain number of rounds of circulating knowledge learning, the comprehensive management platform obtains a temporary neural network. The temporary neural network has certain feature extraction and representation capability, and can provide basic support for subsequent fire risk feature mining and prognosis. Meanwhile, due to the fact that a non-guided knowledge learning strategy is adopted, the temporary neural network can learn more potential knowledge and patterns from a large amount of unlabeled data.

In one possible embodiment, step a112 may include:

Step a1121, performing cyclic knowledge learning on the temporary neural network according to the second power fire safety monitoring log subsequence, and generating the first neural network, wherein each round of cyclic knowledge learning performs the following steps:

and step A1122, loading a second electric power fire safety monitoring log selected from the second electric power fire safety monitoring log subsequence to the temporary neural network to generate corresponding fire risk characteristics.

And step A1123, generating a second error parameter according to the fire risk characteristic and the marked fire risk characteristic corresponding to the second electric power fire safety monitoring log, and performing knowledge learning on the temporary neural network according to the second error parameter.

In this embodiment, in the integrated management platform of the electric power facility monitoring center, after the preliminary training of the non-guided learning strategy, a temporary neural network has been generated. The temporary neural network has certain feature extraction and embedding representation capabilities. Then, the comprehensive management platform performs further parameter learning on the temporary neural network by using the second power fire protection security monitoring log subsequence with the fire risk feature labeling to generate a first neural network capable of performing fire risk feature mining.

In detail, the integrated management platform randomly selects a log record from the second power fire safety monitoring log subsequence, wherein the log record contains various parameters and data related to fire risk, such as temperature, smoke concentration, flame detection and the like. Meanwhile, the record is also accompanied with expert-labeled fire risk feature labels, and the labels are true descriptions and classifications of fire risks in the log.

And the comprehensive management platform loads the selected second electric power fire safety monitoring log into the temporary neural network. This neural network converts the input log data into an internal representation by which fire risk features in the log are captured and identified.

The temporary neural network processes and analyzes the input second electric power fire safety monitoring log, and extracts the characteristics related to the fire risk. These characteristics may be indicators of abnormal temperature rise, excessive smoke concentration, flame detection signals, etc. that reflect the risk of fire. The temporary neural network converts these features into an internal representation that captures the underlying patterns and structures of fire risk in the log. By such an internal representation, the neural network is able to more accurately identify and understand the fire risk features in the input data.

And the comprehensive management platform compares the fire risk characteristics generated by the temporary neural network with the fire risk characteristics marked in the log. And obtaining a second error parameter by calculating the difference or the error between the two. This second error parameter reflects the gap between the current understanding of the fire risk features in the input log by the temporary neural network and the actual situation.

To reduce errors and improve the performance of the neural network, the integrated management platform uses a back-propagation algorithm and a gradient descent optimizer to adjust parameters (e.g., weights and biases) of the temporary neural network. This process is accomplished by continually iterating and updating parameters of the neural network. In each iteration, the comprehensive management platform adjusts parameters of the neural network according to the second error parameters, so that the parameters can be better fit with fire risk characteristics in the input data. Through multiple rounds of iteration and parameter adjustment, the performance of the temporary neural network is gradually improved, and finally a first neural network capable of accurately excavating fire risk features is generated.

In one possible implementation, step a110 may further include:

Step a113, for each first example monitoring feature flow path in the first example monitoring feature flow path sequence, obtaining each monitoring feature node information in one first example monitoring feature flow path and flow state information of each monitoring feature node information, performing sequential encoding on each monitoring feature node information according to the flow state information according to a preset encoding rule, generating an example description section, and obtaining power fire safety application environment information associated with the one first example monitoring feature flow path.

At the power facility monitoring center, the integrated management platform has generated a first neural network through previous training steps, the first neural network having fire risk feature mining capabilities. The integrated management platform will now use the first example monitored feature flow path sequence to further learn parameters of the first neural network in order to generate a second neural network that is capable of extracting feature variation vectors in the monitored feature flow paths in real time.

In detail, the integrated management platform begins processing each of the first example monitoring feature flow paths in the sequence of first example monitoring feature flow paths. These first example monitoring signature flow paths represent flow trajectories and state changes of signature data in an electrical fire safety monitoring system. Each of the first example monitored characteristic flow paths includes a series of monitored characteristic node information and flow state information for the nodes.

The integrated management platform firstly acquires all monitoring feature node information in a first example monitoring feature flow path. Such monitoring feature node information may include the type, location, associated devices, etc. of the node. Meanwhile, the integrated management platform also acquires flow state information corresponding to the monitoring characteristic node information, such as data transmission speed, flow direction, whether abnormality exists or not, and the like.

And the comprehensive management platform performs serialization coding on the obtained monitoring characteristic node information and the flow state information according to a preset coding rule. This serialization encoding process may be to convert node information and flow state information into a particular format or sequence so that the neural network can process and understand. The encoded results are referred to as an example description segment, which is an abstraction and representation of the monitored feature flow path.

In addition to the monitoring feature node information and the flow state information, the integrated management platform also obtains electrical fire safety application environment information associated with the currently processed first example monitoring feature flow path. Such electrical fire safety application environment information may include the operational status of the electrical utility, ambient temperature, humidity, etc., which are critical to understanding and analyzing the changes in the characteristic flow path.

Step a114, performing cyclic knowledge learning on the first neural network according to the description sections of the examples and the corresponding power fire safety application environment information, and generating the second neural network, wherein each round of cyclic knowledge learning performs the following steps:

Step a115, selecting an example description segment from the example description segments, performing initialization feature processing on the selected example description segment according to a preset non-guided learning task, generating a processed example description segment, and constructing guided annotation data corresponding to the example description segment.

And step A116, loading the processed example description section and corresponding electric power fire safety application environment information to the first neural network to generate a corresponding monitoring characteristic flow path prediction result.

And step A117, generating a first error parameter according to the prediction result of the monitoring characteristic flow path and corresponding guiding labeling data, and performing knowledge learning on the first neural network according to the first error parameter.

And the comprehensive management platform performs cyclic knowledge learning on the first neural network by using the acquired example description section and corresponding electric power fire safety application environment information. In each round of circulation, the integrated management platform selects an example description section, and performs initialization feature processing on the example description section to generate processed example description section and corresponding guidance labeling data. And then, the integrated management platform loads the processed example description section and the electric power fire safety application environment information into a first neural network to generate a prediction result of the monitoring characteristic flow path. And finally, generating error parameters according to the prediction result and the instructive annotation data by the comprehensive management platform, and carrying out parameter adjustment and learning on the first neural network by using the error parameters.

Through multiple rounds of cyclic knowledge learning, the performance of the first neural network is gradually improved, and finally a second neural network capable of extracting and monitoring feature change vectors in the feature flow path in real time is generated. The second neural network can be applied to an actual electric power fire safety monitoring system, and is used for analyzing and predicting the characteristic flow path in real time, finding potential fire risks in time and taking corresponding measures.

In one possible embodiment, step a120 may include:

step a121, performing cyclic knowledge learning on the second neural network according to the second example monitoring feature flow path sequence, generating the third neural network, and performing the following steps in each round of cyclic knowledge learning:

step a122, selecting a second example monitoring feature flow path from the second example monitoring feature flow path sequence, and obtaining a feature state change track of the second example monitoring feature flow path.

Step a123, loading the feature state change track of the second exemplary monitoring feature flow path to the second neural network, and generating corresponding potential fire hazard object information, where the potential fire hazard object information includes predicted potential fire hazard objects and hazard index features of the potential fire hazard objects.

And step A124, generating a third error parameter according to the potential fire hazard object information and the labeled potential fire hazard object data corresponding to the second example monitoring characteristic flow path, and performing knowledge learning on the second neural network according to the third error parameter.

At the power facility monitoring center, the integrated management platform has generated a second neural network through the previous training steps, and the network has the capability of extracting the characteristic change vector in the monitoring characteristic flow path in real time. The integrated management platform will now use the second exemplary monitoring feature flow path sequence to further learn parameters of the second neural network in order to generate a third neural network that is capable of extracting potential fire risk objects and their risk indicator features in the monitoring feature flow path.

In detail, the integrated management platform begins cyclic knowledge learning on the second neural network using the second example monitoring feature flow path sequence. In this process, the integrated management platform will constantly take example paths from the sequence and use them to train and optimize the neural network. Through multiple rounds of loop learning, the performance of the second neural network is gradually improved, and finally a third neural network which is more accurate and efficient is generated.

In each round of cyclic knowledge learning, the integrated management platform first randomly selects a path from the second example monitoring characteristic flow path sequence as a current learning sample. This path represents the flow trace and state change of the characteristic data over a period of time in the electrical fire safety monitoring system.

And the comprehensive management platform carries out deep analysis on the selected second example monitoring characteristic flow path and extracts state change information of each characteristic node in the path. Such information may include a trend of variation in characteristic values, occurrence timing and frequency of abnormal fluctuations, and the like. Through comprehensive processing and analysis of the information, the comprehensive management platform generates a complete characteristic state change track description.

The integrated management platform inputs the extracted characteristic state change track data into a second neural network for processing and analysis. The neural network automatically learns and identifies potential modes and structures in the neural network according to the input data, and outputs corresponding potential fire hazard object information. Such information includes predicted potential fire hazard objects and their respective hazard index characteristic values.

And comparing and analyzing the potential fire hazard object information output by the second neural network with the labeling data corresponding to the second example monitoring characteristic flow path by the comprehensive management platform. The integrated management platform generates a third error parameter by calculating a difference or error value between the two. This parameter reflects the gap between the current ability of the second neural network to identify potential fire objects and their hazard index features and the actual situation.

To reduce errors and improve performance of the second neural network, the integrated management platform uses a back-propagation algorithm and a gradient descent optimizer to adjust parameters (e.g., weights and biases) of the second neural network. In each iteration, the comprehensive management platform adjusts the parameter setting of the neural network according to the third error parameter, so that the neural network fits the potential fire hazard objects and the dangerous index characteristics of the potential fire hazard objects in the input data better. Through multiple iterations and parameter adjustments, the performance of the second neural network is gradually improved and a third neural network which is more accurate and efficient is finally generated.

In one possible embodiment, step a130 may include:

Step a131, performing cyclic knowledge learning on the third neural network according to the third example monitoring feature flow path sequence, generating the fire risk pre-judging network, and performing the following steps in each cycle of cyclic knowledge learning:

Step a132, selecting a third example monitoring feature flow path from the third example monitoring feature flow path sequence, and obtaining a feature state change track of the third example monitoring feature flow path and potential fire hazard object information, where the potential fire hazard object information includes each potential fire hazard object and a hazard index feature of each potential fire hazard object related to the third example monitoring feature flow path.

And step A133, loading the characteristic state change track and the potential fire hazard object information of the third example monitoring characteristic flow path to the second neural network to generate a corresponding fire hazard trend evolution graph.

And step A134, generating a fourth error parameter according to the fire risk trend evolution diagram and the marked fire risk trend evolution diagram corresponding to the third example monitoring characteristic flow path, and performing knowledge learning on the third neural network according to the fourth error parameter.

In this embodiment, in the electric power facility monitoring center, the integrated management platform has generated a third neural network through the previous training step, where the third neural network has the capability of extracting the potential fire hazard objects and the dangerous index features thereof in the monitoring feature flow path. The integrated management platform will now use the third exemplary monitoring feature flow path sequence to further learn parameters of the third neural network, in order to generate a neural network capable of fire risk prediction, i.e. a fire risk prediction network.

In detail, the integrated management platform begins cyclic knowledge learning of the third neural network using the third example monitoring feature flow path sequence. In this process, the integrated management platform will continuously select example paths from the third example sequence of monitored feature flow paths and use them to train and optimize the third neural network. Through multiple rounds of cyclic learning, the performance of the third neural network is gradually improved, and finally a neural network with fire risk prediction capability is generated.

In each round of cyclic knowledge learning, the integrated management platform first randomly selects a third example monitoring feature flow path from the third example monitoring feature flow path sequence as a current learning sample. This third example monitoring feature flow path represents the flow trace, state change, and potential fire object information associated with feature data over a period of time in the electrical fire safety monitoring system.

The comprehensive management platform carries out deep analysis on the selected third example monitoring characteristic flow path, and extracts state change information of each characteristic node in the path to form a characteristic state change track. Meanwhile, the comprehensive management platform also acquires potential fire hazard object information related to the path, wherein the potential fire hazard object information comprises types and positions of potential fire hazard objects, dangerous index characteristic values of the potential fire hazard objects and the like.

And the comprehensive management platform takes the extracted characteristic state change track and the potential fire hazard object information as input data, and loads the input data into a third neural network for processing and analysis. The neural network automatically learns and identifies potential modes and structures in the neural network according to the input data, and outputs a corresponding fire hazard trend evolution graph. This figure depicts the risk trend and evolution of a potential fire hazard subject in the future for a period of time.

And the comprehensive management platform compares and analyzes the fire risk trend evolution graph output by the third neural network with the marked fire risk trend evolution graph corresponding to the third example monitoring characteristic flow path. The integrated management platform generates a fourth error parameter by calculating the difference or error value between the two. This fourth error parameter reflects the gap between the current prejudgment capability of the third neural network for the trend of fire risk and the actual situation.

To reduce errors and improve the performance of the neural network, the integrated management platform uses a back-propagation algorithm and a gradient descent optimizer to adjust parameters (e.g., weights and biases) of the third neural network. In each iteration, the comprehensive management platform adjusts the parameter setting of the neural network according to the fourth error parameter, so that the parameter setting is better fitted with the fire risk trend and the evolution process in the input data. Through multiple iterations and parameter adjustment, the performance of the third neural network is gradually improved, and finally a neural network with fire risk pre-judging capability, namely a fire risk pre-judging network is generated.

Fig. 2 illustrates a hardware structure of the integrated management platform 100 for implementing the above-mentioned power fire safety precaution method according to an embodiment of the present application, and as shown in fig. 2, the integrated management platform 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In an alternative embodiment, the integrated management platform 100 may be a single server or a group of servers. The server farm may be centralized or distributed (e.g., integrated management platform 100 may be a distributed system). In an alternative embodiment, the integrated management platform 100 may be local or remote. For example, the integrated management platform 100 may access information and/or data stored in the machine-readable storage medium 120 via a network. For another example, the integrated management platform 100 may be directly connected to the machine-readable storage medium 120 to access stored information and/or data. In an alternative embodiment, the integrated management platform 100 may be implemented on a cloud platform. For example only, the cloud platform may include a private cloud, a public cloud, an aggregated cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any integration thereof.

The machine-readable storage medium 120 may store data and/or instructions. In an alternative embodiment, the machine-readable storage medium 120 may store data acquired from an external terminal. In alternative embodiments, machine-readable storage medium 120 may store data and/or instructions that are used by integrated management platform 100 to perform or use the exemplary methods described herein. In alternative embodiments, machine-readable storage medium 120 may include mass storage, removable storage, volatile read-write memory, read-only memory, and the like, or any integration thereof. Exemplary mass storage devices may include magnetic disks, optical disks, solid state disks, and the like. Exemplary removable memory may include flash drives, floppy disks, optical disks, memory cards, compact disks, tape, and the like.

In a specific implementation, the plurality of processors 110 execute computer executable instructions stored by the machine-readable storage medium 120, so that the processors 110 may execute the power fire safety precaution method according to the above method embodiment, the processors 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processors 110 may be used to describe the transceiving actions of the communication unit 140.

The specific implementation process of the processor 110 may refer to the above-mentioned method embodiments executed by the integrated management platform 100, and the implementation principle and technical effects are similar, which are not described herein again.

In addition, the embodiment of the application also provides a readable storage medium, wherein computer executable instructions are preset in the readable storage medium, and when a processor performs the computer executable instructions, the power fire safety early warning method is realized.

Similarly, it should be noted that in order to simplify the description of the present disclosure and thereby aid in understanding one or more embodiments of the application, various features are sometimes grouped together in a single embodiment, figure, or description thereof.

Claims (10)

1. An electric fire safety pre-warning method, which is characterized by comprising the following steps:

Collecting power fire control monitoring data in a power fire control environment through a sensor network deployed at a key area node, wherein the power fire control monitoring data comprises at least one of temperature, smoke, gas concentration and flame brightness;

Extracting key knowledge features from the power fire control monitoring data according to a predefined power fire control safety monitoring knowledge base, wherein the key knowledge features are used for reflecting the state change of potential fire objects;

Fusing the extracted key knowledge features to generate multi-dimensional feature vector distribution, and generating an instant monitoring feature flow path of the electric fire safety monitoring scene according to dynamic change vectors in the multi-dimensional feature vector distribution;

Inputting an instant monitoring characteristic flow path of an electric power fire safety monitoring scene into a fire risk pre-judging network for completing knowledge learning, and generating a corresponding fire risk trend evolution graph, wherein the instant monitoring characteristic flow path is used for reflecting: the real-time characteristic change vector of each monitoring point in the electric power fire safety monitoring scene is used for reflecting the fire hazard trend evolution graph: and the trend evolution state characteristics of each potential fire hazard object related to the real-time characteristic change vector are obtained.

2. The method of claim 1, wherein the training step of the fire risk prediction network comprises:

According to the electric fire safety monitoring log sequence of the target fire safety tag, parameter learning is carried out on the initialized neural network, a first neural network for carrying out fire risk feature mining of the target fire safety tag is generated, parameter learning is carried out on the first neural network according to the first example monitoring feature flow path sequence, and a second neural network for carrying out real-time feature change vector extraction of the monitoring feature flow path is generated;

Performing parameter learning on the second neural network according to a second example monitoring characteristic flow path sequence to generate a third neural network for extracting potential fire risk objects and dangerous index features of the potential fire risk objects of the monitoring characteristic flow paths, wherein the second example monitoring characteristic flow path sequence comprises each second example monitoring characteristic flow path and corresponding potential fire risk object labeling data;

And performing parameter learning on the third neural network according to a third example monitoring characteristic flow path sequence to generate the fire risk pre-judging network, wherein the third example monitoring characteristic flow path sequence comprises all third example monitoring characteristic flow paths and corresponding marked fire risk trend evolution graphs.

3. The electrical fire safety precaution method of claim 2, wherein the electrical fire safety monitoring log sequence comprises a first electrical fire safety monitoring log sub-sequence and a second electrical fire safety monitoring log sub-sequence;

the parameter learning is performed on the initialized neural network according to the electric fire safety monitoring log sequence of the target fire safety tag, and a first neural network for performing fire risk feature mining of the target fire safety tag is generated, which comprises the following steps:

based on a non-guided knowledge learning strategy, carrying out parameter learning on the initialized neural network according to the first power fire safety monitoring log subsequence to generate a temporary neural network for carrying out fire risk embedding representation;

And according to the second electric fire safety monitoring log subsequence, parameter learning is carried out on the temporary neural network, a first neural network for carrying out fire risk feature mining is generated, and each electric fire safety monitoring log in the second electric fire safety monitoring log subsequence has corresponding fire risk feature labeling.

4. The method of claim 3, wherein the non-guided knowledge learning strategy based on the first power fire safety monitoring log subsequence performs parameter learning on the initialized neural network to generate a temporary neural network for fire risk embedding representation, and the method comprises:

and according to the first power fire safety monitoring log subsequence, performing cyclic knowledge learning on the initialized neural network to generate the temporary neural network, wherein each round of cyclic knowledge learning performs the following steps:

Selecting a first electric fire safety monitoring log from the first electric fire safety monitoring log subsequence, carrying out initialization feature processing on the first electric fire safety monitoring log according to a preset non-guiding learning task, generating a processed first electric fire safety monitoring log, and constructing guiding annotation data corresponding to the first electric fire safety monitoring log;

Loading the processed first electric power fire safety monitoring log to the initialized neural network to generate corresponding fire safety embedded representation data;

generating a first error parameter according to the fire safety embedded representation data and the corresponding guiding annotation data, and performing knowledge learning on the initialized neural network according to the first error parameter.

5. The method of claim 3, wherein the performing parameter learning on the temporary neural network according to the second power fire safety monitoring log sub-sequence to generate a first neural network for performing fire risk feature mining includes:

and according to the second power fire safety monitoring log subsequence, performing cyclic knowledge learning on the temporary neural network to generate the first neural network, wherein each round of cyclic knowledge learning performs the following steps:

Loading a second electric power fire safety monitoring log selected from the second electric power fire safety monitoring log subsequence to the temporary neural network to generate corresponding fire risk characteristics;

Generating a second error parameter according to the fire risk characteristic and the marked fire risk characteristic corresponding to the second electric power fire safety monitoring log, and performing knowledge learning on the temporary neural network according to the second error parameter.

6. The method of claim 2, wherein the performing parameter learning on the first neural network according to the first exemplary monitoring feature flow path sequence to generate a second neural network for performing real-time feature change vector extraction of the monitoring feature flow path comprises:

For each first example monitoring feature flow path in the sequence of first example monitoring feature flow paths, performing the following steps:

Obtaining each monitoring characteristic node information in a first example monitoring characteristic flow path and the flow state information of each monitoring characteristic node information, carrying out serialization coding on each monitoring characteristic node information according to the flow state information according to a preset coding rule to generate an example description section, and obtaining electric fire safety application environment information associated with the first example monitoring characteristic flow path;

According to the example description sections and the corresponding power fire safety application environment information, performing cyclic knowledge learning on the first neural network to generate the second neural network, wherein each round of cyclic knowledge learning performs the following steps:

Selecting an example description section from the example description sections, initializing the selected example description section according to a preset non-guiding learning task, generating a processed example description section, and constructing guiding annotation data corresponding to the example description section;

Loading the processed example description section and corresponding electric power fire safety application environment information to the first neural network to generate a corresponding monitoring characteristic flow path prediction result;

Generating a first error parameter according to the monitoring characteristic flow path prediction result and the corresponding guiding labeling data, and performing knowledge learning on the first neural network according to the first error parameter.

7. The method of any one of claims 2-6, wherein the performing parameter learning on the second neural network according to the second exemplary monitoring feature flow path sequence to generate a third neural network for performing the monitoring feature flow path of the potential fire hazard object and the risk index feature extraction of the potential fire hazard object, includes:

Performing cyclic knowledge learning on the second neural network according to the second example monitoring characteristic flow path sequence to generate the third neural network, wherein each round of cyclic knowledge learning performs the following steps:

selecting a second example monitoring characteristic flow path from the second example monitoring characteristic flow path sequence, and obtaining a characteristic state change track of the second example monitoring characteristic flow path;

loading the characteristic state change track of the second example monitoring characteristic flow path to the second neural network to generate corresponding potential fire hazard object information, wherein the potential fire hazard object information comprises predicted potential fire hazard objects and dangerous index characteristics of the potential fire hazard objects;

And generating a third error parameter according to the potential fire hazard object information and the labeled potential fire hazard object data corresponding to the second example monitoring characteristic flow path, and performing knowledge learning on the second neural network according to the third error parameter.

8. The method of any one of claims 2-6, wherein the performing parameter learning on the third neural network according to a third example monitoring feature flow path sequence to generate the fire risk prediction network includes:

Performing cyclic knowledge learning on the third neural network according to the third example monitoring characteristic flow path sequence to generate the fire risk pre-judging network, wherein each round of cyclic knowledge learning performs the following steps:

selecting a third example monitoring characteristic flow path from the third example monitoring characteristic flow path sequence, and obtaining a characteristic state change track of the third example monitoring characteristic flow path and potential fire hazard object information, wherein the potential fire hazard object information comprises all potential fire hazard objects and dangerous index characteristics of all the potential fire hazard objects related to the third example monitoring characteristic flow path;

loading the characteristic state change track of the third example monitoring characteristic flow path and the potential fire risk object information to the second neural network to generate a corresponding fire risk trend evolution graph;

Generating a fourth error parameter according to the fire risk trend evolution diagram and the marked fire risk trend evolution diagram corresponding to the third example monitoring characteristic flow path, and performing knowledge learning on the third neural network according to the fourth error parameter.

9. The method for electric fire safety pre-warning according to any one of claims 2 to 6, wherein the inputting the real-time monitoring feature flow path of the electric fire safety monitoring scene into the fire risk pre-judging network for completing knowledge learning to generate the corresponding fire trend evolution graph comprises:

Extracting each monitoring characteristic node information and corresponding flow state information in the instant monitoring characteristic flow path through a fire risk pre-judging network, and generating a characteristic state change track of the instant monitoring characteristic flow path according to each monitoring characteristic node information and corresponding flow state information;

and identifying each potential fire hazard object and the dangerous index characteristic of each potential fire hazard object related to the instant monitoring characteristic flow path by using the fire hazard pre-judging network according to the characteristic state change track, and generating a map according to the flow state information of the node information of each monitoring characteristic to generate a corresponding fire hazard trend evolution map.

10. An integrated management platform comprising a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the electrical fire safety precaution method of any one of claims 1-9.