Disclosure of Invention
The invention mainly aims to provide a linkage early warning method based on security situation assessment and application thereof, and solves the problems that the prior art has low efficiency and lacks flexibility when processing complex data, and cannot respond to emergencies in real time.
In order to solve the technical problems, the technical scheme adopted by the invention is that the linkage early warning method based on security situation assessment comprises the following steps:
S1, carrying out normalization, noise filtering and dimension reduction on original data and carrying out feature extraction on the preprocessed data;
S2, constructing an abnormal evaluation model and a normal evaluation model, namely inputting the processed characteristic data into a neural network for training, enabling the model to accurately distinguish between a normal state and an abnormal state by adjusting a network weight and a threshold value, respectively inputting the data in the normal state and the abnormal state into the convolutional neural network model, constructing a safety evaluation model in the normal state and the abnormal state, namely, the normal evaluation model for processing the normal state and the abnormal evaluation model for processing the abnormal state, measuring the difference between the model output and an actual label by adopting a cross entropy loss function, and optimizing the network weight by using an Adam optimizer;
S3, evaluating the safety situation, namely evaluating the data acquired in real time by utilizing a safety evaluation function, judging the current safety state of the hydropower plant, and outputting a safety parameter;
S4, constructing a decision function based on the safety parametersAnd environmental factors EnvFactors by a decision functionDetermining whether the current state is abnormal;
S5, judging the safety state in real time, namely dynamically selecting and determining to select an abnormal evaluation model or a normal evaluation model for real-time evaluation through the step S4, and judging whether the current state needs early warning or not;
S6, linkage early warning, namely after the step S5 judges that early warning is needed, the linkage emergency command system and the early warning information pushing system perform early warning and emergency response, automatically send out early warning signals and prompt the safety risk possibly faced by the hydropower plant;
through the steps, linkage early warning of safety situation assessment is realized.
In a preferred embodiment, the step S1 specifically includes the following substeps:
S11, preprocessing original data, namely carrying out normalization processing on the original data, carrying out noise filtration on the original data by using a Kalman filter, carrying out dimension reduction processing on the data by adopting a principal component analysis technology, and extracting main features in the data;
And S12, extracting features, namely calculating an average value and a standard deviation in a sliding window by utilizing the preprocessed data, extracting features reflecting the running condition of the hydropower plant, analyzing and extracting trend and periodic features by adopting a time sequence, and identifying key frequency components by using a frequency domain analysis method.
In the preferred scheme, in the step S1, the original data comprise water level sensor data, meteorological data, rainfall data and video monitoring data;
Water level sensor data, namely water level data;
Weather data, namely rainfall probability predicted by a weather department;
Rainfall data including rainfall and rainfall duration;
The video monitoring data is used for monitoring the environment around the factory building in real time, and potential risk factors including infrastructure damage, artificial activity risks and equipment faults and maintenance requirements are detected and quantized into video monitoring data.
In a preferred embodiment, in step S2, the processed feature data is input into a neural network for training, and a loss function is used to quantify a difference between the prediction probability and the actual label, where the loss function is specifically expressed as:
;
Wherein,Is a label of whether flooding accident actually occurs, is 0 or 1, the number of N training samples,Representing model predictionThe probability of flooding of each sample,、AndRespectively the firstWater level, rainfall and rainfall probability of each sample.
In a preferred embodiment, in step S2, an Adam optimizer is used to adjust the network weight, where the specific expression of the Adam optimizer is as follows:
;
Wherein,AndIs a modified version of the bias of the first and second moment estimates taking into account the gradients of water level, rainfall and rainfall probability,Representing model parameters, updated continually by the Adam optimizer,Is pointed at the firstThe values of the model parameters at the time of the step,Is pointed at the firstThe values of the model parameters at the time of the step,The learning rate is indicated as being indicative of the learning rate,The water level is indicated and the water level,Indicating the amount of rainfall,Representing the rainfall probability predicted by the meteorological department,Is a numerical value, preventing the denominator from being zero.
In a preferred embodiment, in step S3, the expression of the security evaluation function is as follows:
;
Wherein,The CNN model is represented as such,The parameters of the model are represented by the parameters,The water level is indicated and the water level,Indicating the amount of rainfall,Representing the rainfall probability predicted by the meteorological department.
In a preferred embodiment, in step S4, a decision function is constructedTo judge whether the real-time data is abnormal or not and to decide the functionThe concrete steps are as follows:
;
Based on the security parameters in step S3And environmental factors EnvFactors by a decision functionJudging whether the real-time data is abnormal or not;
Wherein EnvFactors includes environmental factors of time, season and frequency of historical events,Is a parameter of a decision function comprising a weight and a threshold value, a decision functionBased on machine learning or heuristic rules to ensure that the most appropriate model is chosen for evaluation in different situations.
In a preferred embodiment, in step S5, the method dynamically selects and determines to select an abnormal evaluation model or a normal evaluation model for real-time evaluation, which specifically includes:
s51, obtaining whether real-time data is abnormal or not according to the step S4;
s52, if the real-time data is judged to be in an abnormal state, verifying and evaluating by using an abnormal evaluation model;
S53, if the real-time data is judged to be in a normal state, verifying and evaluating by using a normal evaluation model, if a normal result is obtained, judging to terminate, continuously collecting the data, and if an abnormal result is obtained, continuously evaluating by using an abnormal evaluation model;
S54, outputting an abnormal result and carrying out early warning.
The invention also provides a linkage early warning system based on security situation assessment, which comprises a data acquisition layer, a data processing layer and a decision layer;
the data acquisition layer is used for monitoring various operation parameters of the hydropower plant in real time, including water level sensor data, weather forecast of a meteorological department, rainfall data and video monitoring data;
the data processing layer is used for preprocessing, extracting features and identifying modes of the acquired data and providing effective information for the decision layer;
the decision layer is used for evaluating the safety state of the hydropower plant based on the neural network safety situation evaluation model and sending out an early warning signal;
The system is used for realizing a linkage early warning method based on security situation assessment.
In the preferred scheme, the system also comprises a water level monitoring system, a video monitoring system, a rainfall early warning system, a real-time data analysis system, an emergency command system and an early warning information pushing system;
The water level monitoring system is characterized in that a water level sensor is arranged in a factory building, whether water exists in the room or not is monitored in real time, and when the water level exceeds a preset safety threshold value, the system automatically sends out an early warning signal;
The rainfall early warning system collects data of a meteorological department, predicts rainfall conditions in a period of time in the future, and sends out early warning signals when strong rainfall weather is predicted;
the video monitoring system is characterized in that cameras are arranged in important areas inside and outside a factory building, and real-time conditions of the factory building are monitored in real time;
The real-time data analysis system is used for collecting water level, rainfall and video monitoring data, predicting the risk of possible water flooding factory accidents through data analysis and processing, and automatically triggering an early warning signal when the risk reaches a certain degree;
The emergency command system starts an emergency command flow immediately after receiving the early warning signal and notifies taking measures;
The early warning information pushing system pushes early warning information to related personnel in a short message, telephone or APP mode, so that dangerous situations can be known timely.
The invention also provides an application of the linkage early warning method based on the security situation assessment, which is applied to the hydrologic monitoring and disaster prevention fields of hydropower stations, reservoirs and river basin management, and can identify potential flood risks, equipment faults and potential safety hazards in early stage under complex and changeable natural environments and operation conditions by means of efficient integration and intelligent analysis of multi-source monitoring data, so that accurate early warning and quick response to disasters are realized.
The invention also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the linkage early warning method based on the security situation assessment.
The present invention also provides a readable storage medium having stored therein a computer program comprising program code for controlling a process to perform a process comprising a linked early warning method according to a security situation based assessment.
The invention provides a linkage early warning method based on security situation assessment and application thereof, which are used for integrating multisource data such as water level, weather, video monitoring and the like, constructing abnormal assessment and normal assessment models by utilizing a neural network through data preprocessing and feature extraction so as to distinguish normal and abnormal states of a hydropower plant, acquiring operation parameters in real time by a system, judging whether to trigger early warning by a decision layer based on a security assessment function, environmental factors and decision functions after data processing, dynamically selecting an applicable model to carry out verification assessment, and linking an emergency command and early warning pushing system to start response and notify related personnel when confirming abnormality, wherein the whole system covers subsystems such as water level monitoring, rainfall early warning, video monitoring and the like, realizes omnibearing monitoring, intelligent early warning and quick response of security risks of the hydropower plant, remarkably improves security management efficiency, prevents accidents, and ensures production security and personnel life security.
The invention has the beneficial effects that:
(1) The multi-source data fusion and the depth analysis realize the omnibearing monitoring of the running state of the hydropower plant by integrating the information of various sources such as water level sensor data, meteorological data, video monitoring data and the like, perform professional preprocessing and feature extraction on the data, fully utilize the advantages of the multi-element data and improve the comprehensiveness and accuracy of situation awareness;
(2) The accurate safety situation assessment comprises the steps of constructing an abnormal assessment model and a normal assessment model, and accurately distinguishing the normal state and the abnormal state of the hydropower plant by using a neural network technology;
(3) Constructing a decision function based on safety parameters and environmental factors, realizing intelligent judgment of real-time data, identifying potential risks in time, carrying out real-time verification and evaluation by dynamically selecting an evaluation model, ensuring the accuracy and timeliness of early warning, and immediately linking an emergency command system and an early warning information pushing system once the safety parameters and the environmental factors are judged to be abnormal, so that an emergency plan is started quickly and related personnel are notified, and the risk response time is obviously shortened;
(4) Constructing a linkage early warning system comprising a data acquisition layer, a data processing layer, a decision layer, water level monitoring, rainfall early warning, video monitoring, real-time data analysis, emergency command, early warning information pushing and other subsystems, forming a complete closed loop from data acquisition, processing, analysis to decision and response, and effectively improving the integrity and the synergy of the risk management of the hydropower plant;
(5) The application of the invention obviously enhances the identification, early warning and coping capacity of hydropower plants to various safety risks, is beneficial to preventing serious safety accidents, ensures the continuity and stability of production operation, reduces economic loss caused by shutdown, faults or accidents, and protects the life safety of personnel.
Detailed Description
Example 1
As shown in fig. 1-2, a linkage early warning method based on security situation assessment comprises the following steps:
S1, carrying out normalization, noise filtering and dimension reduction on original data and carrying out feature extraction on the preprocessed data;
S2, constructing an abnormal evaluation model and a normal evaluation model, namely inputting the processed characteristic data into a neural network for training, enabling the model to accurately distinguish between a normal state and an abnormal state by adjusting a network weight and a threshold value, respectively inputting the data in the normal state and the abnormal state into the convolutional neural network model, constructing a safety evaluation model in the normal state and the abnormal state, namely, the normal evaluation model for processing the normal state and the abnormal evaluation model for processing the abnormal state, measuring the difference between the model output and an actual label by adopting a cross entropy loss function, and optimizing the network weight by using an Adam optimizer;
S3, evaluating the safety situation, namely evaluating the data acquired in real time by utilizing a safety evaluation function, judging the current safety state of the hydropower plant, and outputting a safety parameter;
S4, constructing a decision function based on the safety parametersAnd environmental factors EnvFactors by a decision functionDetermining whether the current state is abnormal;
S5, judging the safety state in real time, namely dynamically selecting and determining to select an abnormal evaluation model or a normal evaluation model for real-time evaluation through the step S4, and judging whether the current state needs early warning or not;
S6, linkage early warning, namely after the step S5 judges that early warning is needed, the linkage emergency command system and the early warning information pushing system perform early warning and emergency response, automatically send out early warning signals and prompt the safety risk possibly faced by the hydropower plant;
through the steps, linkage early warning of safety situation assessment is realized.
The invention provides a linkage early warning method based on security situation assessment, which integrates multi-source data such as water level, weather, video monitoring and the like, firstly carries out preprocessing and feature extraction on original data, then inputs the processed feature data into a neural network for training, enables a model to accurately distinguish normal states and abnormal states, respectively builds a normal assessment model and an abnormal assessment model by utilizing normal data and abnormal data in a sample library, builds a decision function by utilizing a security assessment function in combination with environmental factors, judges whether real-time data is abnormal, selects a proper model for assessment, accurately distinguishes normal and abnormal states of a hydropower plant, judges whether to trigger early warning, realizes real-time assessment and judgment of the security situation, and links an emergency command and early warning pushing system when confirming the abnormality, starts response and informs related personnel to form a complete closed loop from data acquisition, processing, analysis to decision and response. The method remarkably improves the risk identification, early warning and handling capacity of the hydropower plant, effectively prevents safety accidents, ensures continuous and stable production, reduces economic loss, protects personnel safety and has remarkable social and economic benefits.
In a preferred embodiment, the step S1 specifically includes the following substeps:
S11, preprocessing original data, namely carrying out normalization processing on the original data, carrying out noise filtration on the original data by using a Kalman filter, carrying out dimension reduction processing on the data by adopting a principal component analysis technology, and extracting main features in the data;
And S12, extracting features, namely calculating an average value and a standard deviation in a sliding window by utilizing the preprocessed data, extracting features reflecting the running condition of the hydropower plant, analyzing and extracting trend and periodic features by adopting a time sequence, and identifying key frequency components by using a frequency domain analysis method.
In the step S11, the data with different dimensions or amplitude ranges are unified in scale by utilizing normalization processing, interference caused by unit and size differences among the data is eliminated, convergence and comparison of a neural network model are facilitated, kalman filtering is used, random noise in original data is removed through a recursive filtering algorithm, data quality is improved, focusing on real signals rather than nonsensical fluctuation is ensured during model training, principal Component Analysis (PCA) is to reduce dimensions through linear transformation, extract main feature vectors of the data, compress a data set, simultaneously retain most of information, simplify model input, lighten calculation load and improve model generalization capability.
In the step S12, the average value and standard deviation of the data in a certain time window are calculated through sliding window statistics, the short-term fluctuation characteristic of the running condition of the hydropower plant is captured and reflected to the local stability and abnormal change, the time sequence analysis is utilized to reveal the trend and the periodic mode of the data evolving along with time, which is helpful for understanding the long-term rule and seasonal influence of the hydropower plant operation and providing valuable clues about the future state change for the model, in addition, the frequency domain analysis can identify key frequency components in the data, such as low-frequency trend, medium-high frequency fluctuation and the like, which are often associated with specific running problems or fault modes and help the model distinguish normal states from abnormal states.
In the preferred scheme, in the step S1, the original data comprise water level sensor data, meteorological data, rainfall data and video monitoring data;
Water level sensor data, namely water level data;
Weather data, namely rainfall probability predicted by a weather department;
Rainfall data including rainfall and rainfall duration;
The video monitoring data is used for monitoring the environment around the factory building in real time, and potential risk factors including infrastructure damage, artificial activity risks and equipment faults and maintenance requirements are detected and quantized into video monitoring data.
The system integrates water level sensor data, meteorological data, rainfall data and video monitoring data, achieves multidimensional and omnibearing monitoring of the safety situation of the hydropower plant, ensures comprehensive coverage of risk factors, improves accuracy and sensitivity of risk identification, provides rich and multiple inputs for neural network model training, powerfully supports safety situation assessment and early warning, and improves safety management and emergency capability of the hydropower plant.
In a preferred embodiment, in step S2, the processed feature data is input into a neural network for training, and a loss function is used to quantify a difference between the prediction probability and the actual label, where the loss function is specifically expressed as:
;
Wherein,Is a label of whether flooding accident actually occurs, is 0 or 1, the number of N training samples,Representing model predictionThe probability of flooding of each sample,、AndRespectively the firstWater level, rainfall and rainfall probability of each sample.
The neural network training is carried out on the processed characteristic data by using a specific loss function, and the method has the advantages of target guiding training, focusing flooding risk identification, multi-factor comprehensive consideration, quantitative evaluation of model performance, minimization of loss to approach an ideal prediction state and accurate distinction of flooding accidents in a normal state and an abnormal state, wherein the influence of water level, rainfall and rainfall probability on the flooding risk is considered.
In a preferred embodiment, in step S2, an Adam optimizer is used to adjust the network weight, where the specific expression of the Adam optimizer is as follows:
;
Wherein,AndIs a modified version of the bias of the first and second moment estimates taking into account the gradients of water level, rainfall and rainfall probability,Representing model parameters, updated continually by the Adam optimizer,Is pointed at the firstThe values of the model parameters at the time of the step,Is pointed at the firstThe values of the model parameters at the time of the step,The learning rate is indicated as being indicative of the learning rate,The water level is indicated and the water level,Indicating the amount of rainfall,Representing the rainfall probability predicted by the meteorological department,Is a numerical value, preventing the denominator from being zero.
The Adam optimizer is adopted to adjust the network weight, and has the advantages of self-adapting learning rate, combining the dynamic term and RMSProp, accelerating training and improving convergence efficiency, efficient memory use, only needing to store moving average values, saving resources, being suitable for large-scale network training, stability and robustness, introducing deviation correction term, relieving gradient estimation deviation and avoiding overlarge learning rate in the initial stage of training.
In a preferred embodiment, in step S3, the expression of the security situation function is as follows:
;
Wherein,The CNN model is represented as such,The parameters of the model are represented by the parameters,The water level is indicated and the water level,Indicating the amount of rainfall,Representing the rainfall probability predicted by the meteorological department.
The safety situation function takes the CNN model as a core, combines key data such as water level, rainfall and rainfall probability to realize accurate quantitative evaluation on the current safety state of the hydropower plant, plays a core safety evaluation role in an integral early warning system, performs deep fusion analysis on multiple factors through a deep learning technology, ensures the comprehensiveness and accuracy of evaluation, can dynamically respond to risk change, and outputs safety parameters and environmental parameters to construct a decision functionThereby further selecting a normal evaluation model or an abnormal evaluation model for real-time data evaluation.
In a preferred embodiment, in step S4, a decision function is constructedTo judge whether the real-time data is abnormal or not and to decide the functionThe concrete steps are as follows:
;
Based on the security parameters in step S3And environmental factors EnvFactors by a decision functionJudging whether the real-time data is abnormal or not;
Wherein EnvFactors includes environmental factors of time, season and frequency of historical events,Is a parameter of a decision function comprising a weight and a threshold value, a decision functionBased on machine learning or heuristic rules to ensure that the most appropriate model is chosen for evaluation in different situations.
The decision function plays a key model selection role in the linkage early warning method, dynamically combines real-time safety parameters with environmental factors including time, seasons and historical event frequency, ensures that the early warning system selects the most suitable assessment model, namely the normal or abnormal assessment model, under different running environments and risk conditions, and enhances the environment adaptability and risk identification accuracy of the system. Through dynamic adjustment of the evaluation strategy, the early warning system can flexibly switch between normal/abnormal models, so that not only is the excessive calculation avoided when the risk is low, but also the deep analysis is ensured when the risk is high. In addition, the decision function can optimize resource allocation, so that the operation efficiency of the early warning system is improved, and the stability and response speed are maintained especially when a large amount of real-time data and complex analysis tasks are processed. Based on machine learning or heuristic rules, the decision function can continuously learn and improve model selection accuracy, so that risk identification capacity and prediction accuracy of the early warning system are further improved, and false alarm rate are reduced.
In a preferred embodiment, in step S5, the method dynamically selects and determines to select an abnormal evaluation model or a normal evaluation model for real-time evaluation, which specifically includes:
s51, obtaining whether real-time data is abnormal or not according to the step S4;
s52, if the real-time data is judged to be in an abnormal state, verifying and evaluating by using an abnormal evaluation model;
S53, if the real-time data is judged to be in a normal state, verifying and evaluating by using a normal evaluation model, if a normal result is obtained, judging to terminate, continuously collecting the data, and if an abnormal result is obtained, continuously evaluating by using an abnormal evaluation model;
S54, outputting an abnormal result and carrying out early warning.
The invention also provides a linkage early warning system based on the security situation assessment, which comprises a data acquisition layer, a data processing layer and a decision layer.
And the data acquisition layer is used for monitoring various operation parameters of the hydropower plant in real time, including water level sensor data, weather forecast of a meteorological department, rainfall data and video monitoring data.
And the data processing layer is used for preprocessing, extracting features and identifying modes of the acquired data and providing effective information for the decision layer.
And the decision layer is used for evaluating the safety state of the hydropower plant based on the neural network safety situation evaluation model and sending out an early warning signal.
The system is used for realizing a linkage early warning method based on security situation assessment.
The linkage early warning system comprises a data acquisition layer, a data processing layer and a decision layer, realizes comprehensive monitoring, accurate assessment and timely early warning of the safety situation of the hydropower plant, monitors various operation parameters in real time, covers multi-source data such as water level, weather, rainfall, video monitoring and the like, ensures the comprehensiveness of risk information, improves the data quality through preprocessing, feature extraction and pattern recognition, digs deep risk association, provides effective information for decision making, evaluates the safety situation based on a neural network model, accurately judges the risk level, and timely sends out early warning signals.
In the preferred scheme, the system further comprises a water level monitoring system, a video monitoring system, a rainfall early warning system, a real-time data analysis system, an emergency command system and an early warning information pushing system.
The water level monitoring system is characterized in that a water level sensor is arranged in a factory building, whether water exists in the room or not is monitored in real time, and when the water level exceeds a preset safety threshold, the system automatically sends out an early warning signal.
The rainfall early warning system collects data of a meteorological department, predicts rainfall conditions in a period of time in the future, and sends out early warning signals when strong rainfall weather is predicted.
The video monitoring system is characterized in that cameras are arranged in important areas inside and outside the factory building, and real-time conditions of the factory building are monitored in real time.
And the real-time data analysis system is used for collecting water level, rainfall and video monitoring data, predicting the risk of possible water flooding factory accidents through data analysis and processing, and automatically triggering an early warning signal when the risk reaches a certain degree.
And the emergency command system immediately starts an emergency command flow after receiving the early warning signal and notifies taking measures.
The early warning information pushing system pushes early warning information to related personnel in a short message, telephone or APP mode, so that dangerous situations can be known timely.
The water level monitoring system monitors the water level in the factory in real time, and automatically pre-warns the water level in a super-threshold value and timely discovers the flooding risk. The rainfall early warning system predicts rainfall by using meteorological data, early warns water level rising possibly caused by strong rainfall, and takes precautions in advance. The real-time data analysis system integrates multi-source data, predicts the risk of a flooding accident and accurately triggers early warning. The linkage alarm system realizes the sharing and linkage of internal and external information, expands the early warning range and requests external support. The linkage early warning system realizes comprehensive monitoring, accurate prediction, quick response and effective communication of risks and effectively ensures safe operation of a hydropower plant.
Example 2
1-2, A linkage early warning method based on security situation assessment is as follows:
Step one, constructing a network security situation index system
The original data comprise water level sensor data, meteorological data, rainfall data and video monitoring data.
The CNN predicts the possible water level rising condition by analyzing the trend and abnormal pattern in the sensor data.
Weather data, weather department predicted rainfall probability, weather data, such as predicted rainfall and duration, may also be included in the analysis to evaluate weather conditions that may have an impact on water level.
Rainfall data, including rainfall and rainfall duration, and analyzing the rainfall data, particularly heavy rainfall events, to estimate potential impact on reservoir water level.
And the video monitoring data is used for monitoring the environment around the factory building in real time, detecting the water level change and other potential risk factors and quantifying the water level change and other potential risk factors into video monitoring data.
Step two, constructing a neural network security situation assessment model
(1) And the data preprocessing comprises the steps of carrying out operations such as normalization, filtering and the like on the original data, reducing noise interference, improving the accuracy of evaluation, and carrying out noise filtering on the original data by using a Kalman filter so as to improve the data quality. And the main component analysis (PCA) technology is adopted to carry out dimension reduction treatment on the data, so that the efficiency and accuracy of subsequent treatment are improved.
(2) Feature extraction, namely extracting representative features such as a sliding average value, a standard deviation and the like from the processed data so as to reflect the running condition of the hydropower plant. Time series analysis is employed to extract time-dependent features in the data, such as trends and periodicity, and frequency domain analysis methods are used to identify key frequency components in the data.
(3) And training the neural network, namely inputting the processed characteristic data into the neural network for training, and adjusting the network weight and the threshold value to enable the neural network to correctly identify the normal state and the abnormal state. Convolutional Neural Networks (CNNs) are selected to process and identify complex data patterns.
(4) And (3) constructing an abnormal evaluation model and a normal evaluation model, namely respectively inputting data in a normal state and an abnormal state into a convolutional neural network model, constructing a safety evaluation model in the normal state and the abnormal state, namely, a normal evaluation model for processing the normal state and an abnormal evaluation model for processing the abnormal state, measuring the difference between the model output and an actual label by adopting a cross entropy loss function, and optimizing the network weight by using an Adam optimizer.
In the present invention, convolutional Neural Networks (CNNs) are selected to handle simple and complex data patterns, respectively. CNNs effectively extract features from input data through their multi-layer structure. These layers include a convolution layer, a pooling layer, and a fully-connected layer, each of which performs specific operations on the input data to extract and learn features.
Convolution layers-these layers use multiple small perceptual areas to extract local features of the input data through convolution operations, each convolution kernel focusing on a particular aspect of the captured data, such as edges or texture.
Pooling layer-the pooling layer is used to reduce the spatial dimension of the data while maintaining important information, which helps to reduce the computational effort and prevent overcomplicating.
And the full connection layer is used for integrating the learned characteristics and executing classification or regression tasks after the data passes through a series of rolling and pooling layers.
In the training process, we use a cross entropy loss function to measure the difference between CNN output and actual label, the cross entropy loss function L can be expressed as:
;
Wherein,Is a label of whether flooding accident actually occurs, is 0 or 1, the number of N training samples,Representing the probability of flooding accidents of the ith sample predicted by the model,、AndRespectively the firstWater level, rainfall and rainfall probability of each sample.
In addition, the network weight is adjusted by adopting an Adam optimizer, the Adam optimizer combines the advantages of momentum and RMSprop, is suitable for processing large-scale and non-convex optimization problems, and the specific formula of the Adam optimizer is as follows:
;
Wherein,AndIs a modified version of the bias of the first and second moment estimates taking into account the gradients of water level, rainfall and rainfall probability,Representing model parameters, updated continually by the Adam optimizer,Is pointed at the firstThe values of the model parameters at the time of the step,Is pointed at the firstThe values of the model parameters at the time of the step,The learning rate is indicated as being indicative of the learning rate,The water level is indicated and the water level,Indicating the amount of rainfall,Representing the rainfall probability predicted by the meteorological department,Is a numerical value, preventing the denominator from being zero.
By combining the strong feature extraction capability of CNN, the cross entropy loss function and the efficient optimization strategy of the Adam optimizer, the performance and accuracy of the model in processing complex network security data can be effectively improved.
The convolution layer and the pooling layer act on input data together to extract key features.
Convolutional layer (Conv):
;
Is the input data, here the water levelRainfall amountAnd rainfall probabilityIn a combination of (a) and (b),AndIs the weight and bias parameters of the convolutional layer; representing a convolution operation for extracting features of the input data.
Activation function (ReLU):
;
this is a nonlinear function that increases the expressive power of the model, allowing it to learn more complex features, and the ReLU activation function increases the model's ability to recognize nonlinear relationships.
Pooling layer (Pool):
;
The pooling layer performs dimension reduction operation on Z, reduces the calculated amount and simultaneously reserves important characteristics.
(5) Evaluating safety situation by evaluating real-time collected data with safety evaluation function, judging current safety state of hydropower plant, and outputting a safety parameterRegarding the security state evaluation, the security state evaluation functions are respectively expressed asThe function comprehensively analyzes the water levelRainfall amountAnd rainfall probabilityTo evaluate the security status.
The security assessment function may be expressed as:。
is a sigmoid activation function that maps parameters between 0 and 1, representing the probability of security risk.Comprehensively considering a plurality of input parameters, predicting the safety risk of the hydropower plant in a normal state and in an abnormal state,AndIs the weight and bias of the scoring function,Is the output of the last layer of the CNN model, where,The CNN model is represented as such,Representing parameters of the model, the function outputs a safety parameter reflecting the current safety state of the hydropower plant.
(6) Construction of decision functions based on security parametersAnd environmental factors EnvFactors by a decision functionDetermining whether the current state is abnormal or notThe concrete steps are as follows:
;
This decision function is based on security parametersAnd environmental factors EnvFactors to select an appropriate model. Wherein EnvFactors may include environmental factors such as time, season, historical event frequency, etcIs a parameter of a decision function, possibly including weights and thresholds, for determining whether to use a standard model or a complex model, the decision functionMay be implemented based on machine learning or heuristic rules to ensure that the most appropriate model is chosen for evaluation in different situations.
(7) And (3) judging the safety state in real time, namely dynamically selecting and determining to select an abnormal evaluation model and a normal evaluation model for real-time evaluation through the step S4, and judging whether the current state needs early warning or not.
(8) And (3) linkage early warning, namely after the early warning is judged to be needed through the step S5, the linkage emergency command system and the early warning information pushing system perform early warning and emergency response, and automatically send out early warning signals to prompt the possible safety risks of the hydropower plant.
Through the mechanism, the system can flexibly adjust the evaluation model according to actual conditions, and ensure that accurate and efficient safety evaluation can be carried out under different running states. The response speed of the early warning system is improved, the identification capability of potential risks is enhanced, and the overall safety management level of the hydropower plant is greatly improved.
Step three, the linkage comprehensive system performs early warning
And the water level monitoring system is used for setting a water level sensor in the factory building and monitoring whether water exists in the room in real time. When the water level exceeds a preset safety threshold, the system automatically sends out an early warning signal to be transmitted into the system.
And the rainfall early warning system is used for collecting data of a meteorological department and predicting rainfall conditions in a future period of time. When the heavy rainfall weather is predicted, the system sends out an early warning signal which is transmitted into the system.
And the video monitoring system is used for arranging cameras in important areas inside and outside the factory building and monitoring real-time conditions of the factory building in real time.
And the real-time data analysis system is used for collecting data such as water level, rainfall, video monitoring and the like, predicting the risk of possible water flooding factory accidents through data analysis and processing, and automatically triggering an early warning signal when the risk reaches a certain degree.
And the emergency command system is used for immediately starting an emergency command flow after receiving the early warning signal and notifying relevant departments and personnel to take measures, such as starting drainage equipment, organizing personnel evacuation and the like.
The early warning information pushing system pushes early warning information to related personnel and responsible persons in time through short messages, telephones, APP and the like, so that each person can know dangerous situations in time.
And according to the early warning result, adopting corresponding emergency measures, and the early warning aim is to timely find and process the network security risk. When the early warning is triggered, emergency measures such as strengthening network safety protection, limiting network access authority, performing safety exercise and the like can be adopted according to the early warning result so as to ensure network safety.
Example 3
By further describing the embodiment 1 and the embodiment 2, the application of the linkage early warning method based on the security situation assessment is applied to the field of hydrologic monitoring and disaster prevention of hydropower stations, reservoirs and river basin management, and potential flood risks, equipment faults and potential safety hazards can be recognized in early stages under complex and changeable natural environments and operation conditions through efficient integration and intelligent analysis of multi-source monitoring data, so that accurate early warning and quick response to disasters are realized.
Example 4
Further described in connection with embodiments 1 and 2, an electronic device includes a memory having a computer program stored therein and a processor configured to run the computer program to perform a linked early warning method based on a security posture assessment.
Example 5
Further described in connection with embodiments 1 and 2 is a readable storage medium having stored therein a computer program comprising program code for controlling a process to perform a process comprising a linked warning method based on a security situation assessment.
The above embodiments are only preferred embodiments of the present invention, and should not be construed as limiting the present invention, and the scope of the present invention should be defined by the claims, including the equivalents of the technical features in the claims. I.e., equivalent replacement modifications within the scope of this invention are also within the scope of the invention.