Disclosure of Invention
The embodiment of the application provides an intelligent identification and early warning method and system, which are used for solving the problems of insufficient intelligent identification accuracy and early warning response timeliness in the prior art.
In a first aspect, an embodiment of the present application provides an intelligent recognition and early warning method, including:
receiving real-time environment sensing data, predefining an abnormal pattern library, and carrying out dynamic matching processing on the real-time environment sensing data based on the abnormal pattern library to generate a potential abnormal activity list;
Based on the potential abnormal activity list, a space-time dynamic graph nerve flow algorithm is applied, the time sequence data and the space relevance of abnormal activities in the abnormal activity list are dynamically captured through a continuous time graph rolling mechanism in the space-time dynamic graph nerve flow algorithm, and a feature weighting method is adopted to carry out weighting processing on key space-time feature points in a dynamic capturing result, so that an abnormal behavior trend prediction model is generated;
Based on the abnormal behavior trend prediction model, performing emergency degree classification analysis on risk scores output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy, extracting key features of the risk scores, dynamically setting classification thresholds of the risk scores, performing causal relationship analysis on emergency degree classification analysis results by adopting a causal inference technology, and dynamically adjusting the classification thresholds to generate abnormal behavior emergency degree;
And outputting customized safety notification based on the emergency degree of the abnormal behavior, sending the customized safety notification to a target source group of the real-time environment perception data, continuously monitoring the user behavior and environment change of the target source group, dynamically updating the abnormal pattern library, and generating an intelligent recognition and early warning strategy.
Optionally, based on the potential abnormal activity list, a spatiotemporal dynamic graph neural flow algorithm is applied, dynamic capturing is performed on time sequence data and spatial relevance of abnormal activities in the abnormal activity list through a continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, a feature weighting method is adopted to perform weighting processing on key spatiotemporal feature points in a dynamic capturing result, and an abnormal behavior trend prediction model is generated, including:
based on the potential abnormal activity list, collecting heterogeneous data in the potential abnormal activity list by adopting a multi-source information fusion technology to carry out integration and synchronization processing, and generating a heterogeneous data set;
Based on the heterogeneous data set, a space-time dynamic graph nerve flow algorithm is applied, and the time sequence data of abnormal activities in the abnormal activity list and the spatial relevance are dynamically captured through a continuous time graph convolution mechanism in the space-time dynamic graph nerve flow algorithm, so that an abnormal behavior characteristic map is generated;
based on the abnormal behavior feature map, carrying out weighting treatment on key space-time feature points in the dynamic capturing result by adopting a feature weighting method, and identifying feature points with important influence in the weighting treatment result to generate a key space-time feature set;
Based on the key space-time feature set, modeling the time sequence characteristics of the abnormal behaviors in the abnormal behavior feature map by combining the modeling capability of the space-time dynamic graph nerve flow algorithm, and generating an abnormal behavior trend prediction model.
Optionally, based on the heterogeneous data set, a spatiotemporal dynamic graph neural flow algorithm is applied, and the time series data and the spatial correlation of the abnormal activities in the abnormal activity list are dynamically captured through a continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, so as to generate an abnormal behavior feature map, which includes:
based on the heterogeneous data set, extracting time sequence data and space relevance information in the heterogeneous data set by adopting a data cleaning and denoising technology, and preprocessing to generate a preprocessed data set;
Based on the preprocessing data set, a space-time dynamic graph nerve flow algorithm is applied, and a continuous time graph rolling mechanism in the space-time dynamic graph nerve flow algorithm is used for dynamically capturing time sequence data of abnormal activities in the abnormal activity list and space relevance, so that a dynamic capturing result is generated;
Based on the dynamic capturing result, a pattern recognition technology is applied to classify and cluster abnormal patterns in the dynamic capturing result, and an abnormal pattern library is generated;
Based on the abnormal pattern library, a multidimensional icon is constructed to carry out visual analysis on abnormal behaviors in the abnormal pattern library, the space-time distribution characteristics of the abnormal behaviors are displayed, and an abnormal behavior characteristic map is generated.
Optionally, based on the abnormal behavior feature map, a feature weighting method is adopted to perform weighting processing on key space-time feature points in the dynamic capturing result, and identify feature points with important influence in the weighted processing result, so as to generate a key space-time feature set, including:
Based on the abnormal behavior feature map, extracting representative space-time feature points in the abnormal behavior feature map by a feature point extraction method to generate a preliminary feature point set;
based on the preliminary feature point set, weighting key space-time feature points in a dynamic capturing result by adopting a feature weighting method, and giving different weights to different feature points in the key space-time feature points so as to distinguish the importance degrees of the different feature points and generate a weighted feature point set;
based on the weighted feature point set, performing internal relation extraction processing on the space-time features of different weighted feature points in the weighted feature point set by using an association rule mining technology to generate a space-time association rule base;
And extracting relevant space-time characteristic points in the space-time association rule base based on the space-time association rule base, and carrying out characteristic point combination processing to generate a key space-time characteristic set.
Optionally, the performing emergency degree classification analysis on the risk score output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy based on the abnormal behavior trend prediction model, extracting key features of the risk score, dynamically setting a classification threshold of the risk score, performing causal relationship analysis on an emergency degree classification analysis result by adopting a causal inference technology, dynamically adjusting the classification threshold, and generating the abnormal behavior emergency degree includes:
based on the abnormal behavior trend prediction model, carrying out interval distribution analysis on risk scores output by the abnormal behavior trend prediction model, and evaluating risk probabilities of different scoring intervals in the interval distribution analysis to generate a risk probability distribution map;
Based on the risk probability distribution map, performing emergency grading analysis on the risk scores output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy, extracting key features of the risk scores, dynamically setting grading thresholds of the risk scores, and generating a preliminary emergency grading result;
Based on the preliminary emergency grading result, carrying out causal relationship analysis on an emergency grading analysis result by adopting a causal inference technology, dynamically adjusting the grading threshold value, and generating an optimized emergency grading result;
and establishing a multi-level feedback mechanism based on the optimized emergency degree grading result, setting different feedback paths for emergency degrees of different levels in the optimized emergency degree grading result, and generating abnormal behavior emergency degrees.
Optionally, based on the risk probability distribution diagram, performing emergency grading analysis on the risk score output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy, extracting key features of the risk score, dynamically setting a grading threshold of the risk score, and generating a preliminary emergency grading result, including:
deep analysis is carried out based on the risk probability distribution diagram, the overall risk mode and trend in the risk probability distribution diagram are focused, vectorization is carried out on the overall risk mode and trend, and a risk feature matrix is generated;
Based on the risk feature matrix, applying a federal transfer learning algorithm based on differential privacy to each risk feature vector in the risk feature matrix, and carrying out emergency grading analysis on risk scores output by the abnormal behavior trend prediction model to generate emergency grading scores;
Based on the emergency grading score, carrying out score fusion processing by extracting key features of the risk score, dynamically setting a grading threshold of the risk score according to a score fusion processing result, and generating an optimized grading threshold;
And introducing an adaptive adjustment mechanism based on the optimized grading threshold, recording output data in real time during the running of the abnormal behavior trend prediction model, and comparing and analyzing with the optimized grading threshold to automatically adjust the output data to generate a preliminary emergency grading result.
Optionally, the outputting the customized security notification based on the emergency degree of the abnormal behavior is sent to a target source group of the real-time environment sensing data, continuously monitoring the user behavior and the environment change of the target source group, dynamically updating the abnormal pattern library, and generating the intelligent recognition and early warning strategy, including:
Outputting customized safety notification based on the emergency degree of the abnormal behavior, and sending the customized safety notification to a target source group of the real-time environment perception data to generate a customized safety notification record;
Setting an automatic response flow based on the customized security notification record so as to automate the response flow of the target source group after receiving the customized security notification record and generate an automatic response log;
Continuously monitoring user behaviors and environmental changes of the target source group based on the automatic response log, capturing potential abnormal activities in the continuous monitoring process, and generating a real-time monitoring report;
And extracting a newly generated abnormal mode in the real-time monitoring report based on the real-time monitoring report so as to dynamically update the abnormal mode library and generate an intelligent recognition and early warning strategy.
In a second aspect, an embodiment of the present application provides an intelligent recognition and early warning system, including:
the receiving module is used for receiving the real-time environment sensing data, predefining an abnormal pattern library, and carrying out dynamic matching processing on the real-time environment sensing data based on the abnormal pattern library to generate a potential abnormal activity list;
the processing module is used for dynamically capturing the time sequence data and the space relevance of the abnormal activities in the abnormal activity list by using a continuous time graph convolution mechanism in the space-time dynamic graph nerve flow algorithm based on the potential abnormal activity list and applying a space-time dynamic graph nerve flow algorithm, and carrying out weighting processing on key space-time characteristic points in a dynamic capturing result by adopting a characteristic weighting method to generate an abnormal behavior trend prediction model;
The analysis module is used for carrying out emergency degree grading analysis on risk scores output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy, extracting key characteristics of the risk scores, dynamically setting grading thresholds of the risk scores, carrying out causal relation analysis on emergency degree grading analysis results by adopting a causal inference technology, and dynamically adjusting the grading thresholds to generate abnormal behavior emergency degree;
The monitoring module is used for outputting customized safety notification based on the emergency degree of the abnormal behavior, sending the customized safety notification to a target source group of the real-time environment perception data, continuously monitoring the user behavior and environment change of the target source group, dynamically updating the abnormal pattern library and generating an intelligent recognition and early warning strategy.
In a third aspect, an embodiment of the present application provides a computing device, including a processing component and a storage component, where the storage component stores one or more computer instructions, and the one or more computer instructions are used to be invoked and executed by the processing component to implement an intelligent recognition and early warning method according to the first aspect.
In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, where the computer program when executed by a computer implements an intelligent recognition and early warning method as described in the first aspect.
In the embodiment of the application, real-time environment perception data are received, an abnormal pattern library is predefined, dynamic matching processing is carried out on the real-time environment perception data based on the abnormal pattern library, a potential abnormal activity list is generated, a time sequence data and space relativity of abnormal activities in the abnormal activity list are dynamically captured through a continuous time graph convolution mechanism in the time-space dynamic graph nerve flow algorithm based on the potential abnormal activity list, a feature weighting method is adopted, key time-space feature points in a dynamic capturing result are weighted to generate an abnormal behavior trend prediction model, emergency degree classification analysis is carried out on risk scores output by the abnormal behavior trend prediction model based on the abnormal behavior trend prediction model by utilizing a federal migration learning algorithm based on differential privacy, key features of the risk scores are extracted, classification thresholds of the risk scores are dynamically set, causal inference technology is adopted, causal relation analysis is carried out on emergency degree classification analysis results, the classification thresholds are dynamically adjusted, abnormal behavior degrees are generated, customized safety notification is output based on the abnormal behavior degree, the abnormal behavior trend prediction model is sent to the target environment perception model, a new user behavior is identified with a target group, and a new dynamic environment is generated. The method comprises the steps of dynamically matching real-time environment perception data with a predefined abnormal pattern library to ensure accurate capture of potential abnormal activities, adopting a space-time dynamic image nerve flow algorithm to effectively process complex space-time correlation and improve adaptability to different environment changes, outputting customized safety notification based on emergency degree of abnormal behaviors to realize accurate early warning for specific groups and improve response speed and effectiveness, continuously monitoring behaviors and environment changes of target source groups, dynamically updating the abnormal pattern library to ensure self-adaptability and long-term reliability of a system, and introducing differential privacy technology in an analysis process to ensure safety and privacy of user data.
Further, heterogeneous data are integrated through a multi-source information fusion technology to generate a comprehensive and consistent heterogeneous data set, usability and accuracy of the data are enhanced, key space-time characteristic points are identified by adopting a characteristic weighting method, influence of important characteristics is highlighted, interpretation ability and prediction accuracy of a model are improved, modeling ability of a space-time dynamic graph nerve flow algorithm is combined to generate an abnormal behavior trend prediction model, scientific basis is provided for decision making, and reliability and practicality of prediction are enhanced.
Further, the risk scores are subjected to interval distribution analysis, risk probabilities of different scoring intervals are evaluated, an accurate risk probability distribution map is generated, a solid foundation is provided for subsequent analysis, a federal migration learning algorithm based on differential privacy is used for emergency classification analysis, a causal inference technology is combined for dynamically adjusting a threshold value, scientificity and rationality of classification results are ensured, a multi-stage feedback mechanism is established, different feedback paths are set for emergency degrees of different levels, flexibility and response speed of the system are improved, early warning effectiveness is ensured, a differential privacy technology is applied in the whole process, user privacy in the data analysis process is ensured not to be leaked, and user trust is enhanced.
These and other aspects of the application will be more readily apparent from the following description of the embodiments.
Detailed Description
In order to enable those skilled in the art to better understand the present application, the following description will make clear and complete descriptions of the technical solutions according to the embodiments of the present application with reference to the accompanying drawings.
In some of the flows described in the specification and claims of the present application and in the foregoing figures, a plurality of operations occurring in a particular order are included, but it should be understood that the operations may be performed out of order or performed in parallel, with the order of operations such as 101, 102, etc., being merely used to distinguish between the various operations, the order of the operations themselves not representing any order of execution. In addition, the flows may include more or fewer operations, and the operations may be performed sequentially or in parallel. It should be noted that, the descriptions of "first" and "second" herein are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, and are not limited to the "first" and the "second" being different types.
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the application.
Fig. 1 is a flowchart of an intelligent recognition and early warning method according to an embodiment of the present application, as shown in fig. 1, the method includes:
101. receiving real-time environment sensing data, predefining an abnormal pattern library, and carrying out dynamic matching processing on the real-time environment sensing data based on the abnormal pattern library to generate a potential abnormal activity list;
In this step, the real-time environmental perception data refers to environmental information, such as temperature, humidity, illumination intensity, object movement track, etc., collected by a sensor, a camera, etc., during a specific period of time.
The anomaly pattern library is a pre-built database containing various known anomaly patterns and features for matching with real-time perceptual data to identify potential anomaly activity.
The dynamic matching process refers to a process of comparing real-time environment sensing data with patterns in an abnormal pattern library, and the similarity between the real-time environment sensing data and the patterns is automatically analyzed through an algorithm to determine whether potential abnormal activities exist.
The potential abnormal activity list is a list generated after dynamic matching processing, and lists all possible abnormal activities and relevant information thereof, including occurrence time, place, type and the like.
In the embodiment of the application, firstly, a system receives real-time environment sensing data from a plurality of sensors and monitoring equipment, secondly, the system loads a predefined abnormal pattern library to prepare for matching, thirdly, the real-time data is compared with the abnormal pattern by using a dynamic matching algorithm to calculate the similarity, and finally, a potential abnormal activity list is generated according to the matching result and all suspected abnormal events are recorded.
It is assumed that in an intelligent post-emergency delivery pod, the system needs to monitor the real-time status of the patient and provide intelligent adjuvant therapy. Firstly, a camera is arranged in a cabin and used for monitoring the state of a patient in an unmanned emergency transport process, data analysis is carried out through information such as facial expression, limb behavior and the like of the patient, secondly, a system loads a mode library containing various abnormal physiological states (such as shortness of breath, confusion, limb twitches and the like), thirdly, a dynamic matching algorithm is used for comparing real-time data with a predefined mode, similarity scores are calculated, potential emergency medical conditions are identified, and finally, a potential abnormal activity list is generated and all possible abnormal conditions such as dyspnea or loss of consciousness of the patient suddenly appear.
102. Based on the potential abnormal activity list, a space-time dynamic graph nerve flow algorithm is applied, the time sequence data and the space relevance of abnormal activities in the abnormal activity list are dynamically captured through a continuous time graph rolling mechanism in the space-time dynamic graph nerve flow algorithm, and a feature weighting method is adopted to carry out weighting processing on key space-time feature points in a dynamic capturing result, so that an abnormal behavior trend prediction model is generated;
The space-time dynamic graph nerve flow algorithm is a deep learning algorithm combining space-time dimension and graph structure, can process complex data with time and space correlation, and captures dynamic changes among nodes through a continuous time graph rolling mechanism.
The continuous-time graph rolling mechanism refers to the introduction of a time dimension in the graph neural network, so that the model can capture the relationship changes between nodes at different points in time, which helps to better understand the dynamic patterns in the time series data.
Key spatiotemporal feature points refer to those data points that have significant impact in both temporal and spatial dimensions, which are often important markers of abnormal behavior, critical to subsequent analysis.
The abnormal behavior trend prediction model is a mathematical model generated based on a space-time dynamic graph nerve flow algorithm and is used for predicting the development trend of abnormal behaviors, and the model can help to identify potential risks in advance and provide support for decision making.
In the embodiment of the application, firstly, a system processes by applying a space-time dynamic graph nerve flow algorithm based on data in a potential abnormal activity list, secondly, captures time sequence data and space relevance of abnormal activities through a continuous time graph convolution mechanism, thirdly, adopts a feature weighting method to evaluate and weight key space-time feature points, and finally, generates an abnormal behavior trend prediction model for predicting future evolution of abnormal behaviors.
For example, continuing the above example, assume that in an intelligent emergency back-off pod, the system has generated a list of potential abnormal activities that include abnormal physiological states (e.g., shortness of breath, confusion, etc.) for certain patients. The system firstly processes related data by using a space-time dynamic image nerve flow algorithm based on the list, secondly captures physiological change modes and spatial distribution (such as facial expression and limb actions) of the patients in the transportation process through a continuous time image convolution mechanism, thirdly identifies key space-time characteristic points such as severe respiratory fluctuation or sudden limb twitch of the patients at a certain moment, carries out weighted processing on the key space-time characteristic points, and finally generates an abnormal behavior trend prediction model to predict the possible health deterioration condition of the patients in the future and assist in guiding the improvement of the oxygen concentration of the breathing machine to enhance the comfort, adjust the ventilation amount of tidal volume minutes and other breathing states.
103. Based on the abnormal behavior trend prediction model, performing emergency degree classification analysis on risk scores output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy, extracting key features of the risk scores, dynamically setting classification thresholds of the risk scores, performing causal relationship analysis on emergency degree classification analysis results by adopting a causal inference technology, and dynamically adjusting the classification thresholds to generate abnormal behavior emergency degree;
In this step, the differential privacy technique is a technique for protecting personal privacy, and ensures that specific information of an individual is not revealed in the data analysis process. The effect of a single data point is blurred by adding noise or other means.
The federal transfer learning algorithm is a distributed machine learning method, which allows different participants to train models together without sharing original data, and the transfer learning can utilize knowledge of existing models to accelerate the learning process of new tasks.
The risk score refers to a value representing the risk degree by quantitatively evaluating the result output by the abnormal behavior trend prediction model.
Emergency level analysis is to divide abnormal behavior into different emergency levels according to risk scores so as to take corresponding countermeasures.
Causal inference techniques are a statistical method for analyzing causal relationships between variables. The reasons behind the abnormal behavior can be better understood by this method, thereby developing a more effective intervention strategy.
The emergency degree of the abnormal behavior refers to final risk assessment of the abnormal behavior according to the emergency degree grading analysis result, and the emergency degree of the abnormal behavior is determined.
In the embodiment of the application, firstly, a system performs interval distribution analysis based on risk scores output by an abnormal behavior trend prediction model, secondly, performs emergency degree classification analysis on the risk scores by using a federal migration learning algorithm based on differential privacy, thirdly, extracts key features of the risk scores, dynamically sets classification threshold values, and finally, performs causal relation analysis on classification results by using a causal inference technology, dynamically adjusts the classification threshold values, and generates abnormal behavior emergency degree.
For example, continuing the example above, assume that the system has generated an abnormal behavioral trend prediction model that predicts a future likely health deterioration of a patient. The system firstly carries out interval distribution analysis based on risk scores output by the model to evaluate risk probabilities of different scoring intervals, secondly carries out emergency grading analysis on the risk scores by using a federal migration learning algorithm based on differential privacy to ensure that the privacy of a patient is not violated, thirdly, extracts key features and dynamically sets grading thresholds to divide the risk scores into different emergency grades, and finally carries out causality analysis on grading results by adopting a causality inference technology to dynamically adjust the grading thresholds to generate final abnormal behavior emergency degree and guide medical teams to take proper treatment measures, such as sedative bolus injection, epinephrine bolus injection if necessary and the like for manic patients.
104. And outputting customized safety notification based on the emergency degree of the abnormal behavior, sending the customized safety notification to a target source group of the real-time environment perception data, continuously monitoring the user behavior and environment change of the target source group, dynamically updating the abnormal pattern library, and generating an intelligent recognition and early warning strategy.
In this step, the customized security notification is based on the degree of emergency of the abnormal behavior, and the system generates security pre-warning information for a specific group or scene.
Target source groups refer to those persons or units associated with abnormal behavior, such as affected residents, businesses, government agencies, etc., who are the primary recipients of customized security notifications.
The continuous monitoring means that after the system sends out the safety notification, the system continuously monitors the behavior change and the environmental condition of the target source group so as to evaluate the early warning effect and adjust the strategy in time.
The intelligent recognition and early warning strategy is to dynamically update an abnormal pattern library according to the continuous monitoring result, optimize the recognition capability and early warning mechanism of the system, and ensure that the system can adapt to new threats and changes and keep long-term effectiveness.
In the embodiment of the application, firstly, a system generates customized safety notification based on the emergency degree of abnormal behaviors, secondly, the notification is sent to a target source group of real-time environment perception data, thirdly, the user behaviors and environment changes of the target source group are continuously monitored, and finally, an abnormal pattern library is dynamically updated to generate an intelligent recognition and early warning strategy so as to ensure the self-adaptability and the reliability of the system.
Continuing the example above, assume that the system has generated an abnormal behavioral urgency indicating that a patient is about to experience serious health deterioration. The system firstly generates customized safety notification based on the emergency degree to inform the accompanying medical staff and the remote medical center of paying attention to intensive care, secondly sends the notification to related medical management departments and rescue teams participating in transportation, thirdly continuously monitors the behavior changes and the environmental conditions of the groups to evaluate the early warning effect, such as monitoring the facial expression and limb actions of a patient in real time through a camera, and finally dynamically updates an abnormal mode library according to the monitoring result to generate an intelligent recognition and early warning strategy to ensure that the system can better cope with future similar conditions, thereby improving the emergency effect and the safety of the patient. In summary, steps 101 to 104 cover the complete flow from the receiving of the real-time environmental awareness data and the abnormal pattern matching, to the dynamic capturing and trend prediction of the abnormal behavior, and to the emergency grading analysis of the risk score, until the customized safety notification is finally output and the pre-warning strategy is continuously optimized, so as to provide a comprehensive and intelligent abnormal behavior recognition and pre-warning system, and meet the requirements of public safety and management efficiency in the smart city and the internet of things environment.
In order to further improve accuracy and comprehensiveness of abnormal behavior identification, in some embodiments, the step 102 of applying a spatiotemporal dynamic graph nerve flow algorithm based on a potential abnormal activity list includes collecting heterogeneous data in the potential abnormal activity list based on the potential abnormal activity list by adopting a multi-source information fusion technology, integrating and synchronously processing the heterogeneous data to generate a heterogeneous data set, applying a spatiotemporal dynamic graph nerve flow algorithm based on the heterogeneous data set, dynamically capturing time sequence data and spatial correlation of abnormal activities in the abnormal activity list through a continuous time graph convolution mechanism in the spatiotemporal dynamic graph nerve flow algorithm to generate an abnormal behavior feature map, weighting key spatiotemporal feature points in a dynamic capturing result based on the abnormal behavior feature map by adopting a feature weighting method, identifying feature points with important influences in a weighting processing result to generate a key spatiotemporal feature set, modeling abnormal behavior sequence characteristics in the abnormal behavior feature map based on the key spatiotemporal feature set, and combining capability of the spatiotemporal dynamic graph nerve flow algorithm to generate an abnormal behavior trend prediction model.
In this embodiment, the multi-source information fusion technique refers to a technique of integrating and synchronizing data (such as sensor data, video monitoring data, etc.) from different sources, and these heterogeneous data are cleaned and converted to form a unified data set for subsequent analysis.
Heterogeneous data sets are data sets composed of a plurality of different types of data, such as time-series data, image data, text data, and the like. The data may come from different acquisition devices or systems, with different formats and structures.
The abnormal behavior feature map is a visualization tool generated by a space-time dynamic map nerve flow algorithm, and shows the space-time relationship among feature points in abnormal activities. It helps to intuitively understand the progress and spatial distribution of abnormal behavior.
Key spatiotemporal feature points are those data points in the abnormal behavior feature pattern that are critical to the recognition and prediction of abnormal behavior, which are typically located at critical locations of abnormal behavior and exhibit significant changes in the temporal and spatial dimensions.
The key space-time feature set is a group of feature points with the most representation and influence selected from all key space-time feature points, and the recognition capability and the prediction accuracy of the model can be enhanced through weighting the feature points.
In the embodiment of the application, firstly, a system collects and integrates heterogeneous data by adopting a multi-source information fusion technology based on data in a potential abnormal activity list to generate a comprehensive heterogeneous data set, secondly, a time sequence data and space relevance in the heterogeneous data set are dynamically captured by utilizing a space-time dynamic graph nerve flow algorithm to generate a detailed abnormal behavior feature map, thirdly, importance evaluation is carried out on key space-time feature points in the feature map by adopting a feature weighting method to identify feature points with great influence to generate a key space-time feature set, and finally, modeling is carried out on the time sequence characteristics of abnormal behaviors by combining the powerful modeling capability of the space-time dynamic graph nerve flow algorithm to generate an accurate abnormal behavior trend prediction model.
The following is a specific example:
It is assumed that in an intelligent post-emergency delivery module, the system needs to improve the ability to identify and pre-warn abnormal physiological conditions of the patient. Firstly, a camera is arranged in a cabin and used for monitoring the state of a patient in an unmanned emergency transport process, data analysis is carried out through information such as facial expression, limb behavior and the like of the patient, secondly, a multi-source information fusion technology is adopted in a system, data from a plurality of medical devices such as a breathing machine, an electrocardiograph monitor, an injection pump and the like are collected to generate an isomerism data set containing information such as respiratory rate, heart rate, oxygen saturation, medicine dosage and the like, thirdly, a space-time dynamic image nerve flow algorithm is utilized for processing the data to generate a detailed abnormal behavior characteristic map, physiological change modes and spatial distribution of the patient in the transport process are displayed, and finally, key space-time characteristic points such as severe respiratory fluctuation or sudden limb twitches of the patient at a certain moment are identified and are weighted to generate a key space-time characteristic set.
In order to further improve the accuracy and data quality of abnormal behavior identification, in some embodiments, the step 102 of applying a space-time dynamic graph nerve flow algorithm based on the heterogeneous data set includes extracting time series data and space relevance information in the heterogeneous data set to perform preprocessing based on the heterogeneous data set by adopting a data cleaning and denoising technology to generate a preprocessed data set, applying a space-time dynamic graph nerve flow algorithm based on the preprocessed data set to dynamically capture time series data and space relevance of abnormal activities in the abnormal activity list by using a continuous time graph convolution mechanism in the space-time dynamic graph nerve flow algorithm to generate a dynamic capture result, applying a pattern recognition technology to classify and cluster abnormal patterns in the dynamic capture result based on the dynamic capture result to generate an abnormal pattern library, and performing visual analysis on abnormal behaviors in the abnormal pattern library by constructing a multi-dimensional icon based on the abnormal pattern library to generate abnormal behavior characteristics.
In this embodiment, the data cleaning and denoising technique refers to removing noise and incomplete information in data by a series of processing methods, and ensuring accuracy and consistency of the data, where common methods include filtering, smoothing, missing value filling, and the like.
The preprocessing data set is the data set after cleaning and denoising, so that the core characteristics of the original data are reserved, and meanwhile, the interference factors are reduced.
Time series data refers to data points arranged in time series, and is generally used to describe the change of a certain variable with time.
Spatial association information refers to the interrelationship between different sites, such as the traffic flow effect between two adjacent intersections.
The dynamic capturing result is a result generated after capturing the time sequence data and the spatial correlation of the abnormal activity through a space-time dynamic graph nerve flow algorithm, and reflects the development process of the abnormal behavior and the space-time characteristics thereof.
The pattern recognition technology is a technology for automatically finding rules from data and classifying the rules, and is widely applied to the fields of image recognition, voice recognition and the like.
The abnormal pattern library is a database storing various known abnormal patterns, and each pattern comprises a characteristic description and a classification label.
The multi-dimensional icon is a visualization tool, and can graphically display multiple dimensions of complex data, so that a user can intuitively understand the relevance and the change trend among the data.
The abnormal behavior feature map is formed by carrying out visual analysis on abnormal behaviors in an abnormal pattern library, and the generated chart shows the space-time distribution characteristic of the abnormal behaviors and provides support for decision making.
In the embodiment of the application, firstly, a system adopts a data cleaning and denoising technology based on a heterogeneous data set, extracts time sequence data and space correlation information in the time sequence data and space correlation information to perform preprocessing to generate a high-quality preprocessed data set, secondly, utilizes a continuous time graph convolution mechanism in a space-time dynamic graph nerve flow algorithm to dynamically capture abnormal activity time sequence data and space correlation in the preprocessed data set to generate a detailed dynamic capturing result, thirdly, adopts a pattern recognition technology to classify and cluster abnormal patterns in the dynamic capturing result to construct a comprehensive abnormal pattern library, and finally, performs visual analysis on abnormal behaviors in the abnormal pattern library by constructing a multi-dimensional icon to display the space-time distribution characteristics of the behaviors and generate an abnormal behavior characteristic map.
The following is a specific example:
It is assumed that in an intelligent emergency transport pod, the system needs to promote the ability to identify potential health risks. Firstly, a camera and a plurality of medical sensors are arranged in a cabin and used for monitoring the state of a patient in an unmanned emergency transportation process, data analysis is carried out through information such as facial expressions, limb behaviors and the like of the patient, secondly, a system adopts a data cleaning and denoising technology based on collected heterogeneous data sets (such as respiratory frequency, heart rate, oxygen saturation and medicine dosage) to extract time sequence data and space correlation information in the heterogeneous data sets so as to carry out preprocessing, a preprocessing data set is generated, a continuous time graph rolling mechanism in a space-time dynamic graph nerve flow algorithm is used for carrying out dynamic capture on the abnormal physiological activity time sequence data and the space correlation in the preprocessing data set so as to generate detailed dynamic capture results, then, a mode recognition technology is used for carrying out classification and clustering processing on abnormal modes in the dynamic capture results so as to construct an abnormal mode library containing a plurality of abnormal physiological states, finally, a multidimensional graph is constructed for carrying out visual analysis on abnormal behaviors in the abnormal mode library, the time-space distribution characteristics of the behaviors are displayed, and visual early warning information and support are provided for medical staff.
In order to further improve the accuracy of abnormal behavior identification and the importance distinction of feature points, in some embodiments, a feature weighting method is adopted in the feature map based on abnormal behavior in step 102, and the method comprises the steps of extracting representative spatiotemporal feature points in the abnormal behavior feature map through a feature point extraction method based on the abnormal behavior feature map to generate a preliminary feature point set, weighting key spatiotemporal feature points in a dynamic capturing result by adopting a feature weighting method based on the preliminary feature point set, giving different weights to different feature points in the key spatiotemporal feature points to distinguish the importance degree of the different feature points to generate a weighted feature point set, extracting the spatiotemporal features of the different weighted feature points in the weighted feature point set through an association rule mining technology based on the weighted feature point set to generate a spatiotemporal association rule base, extracting relevant spatiotemporal feature points in the spatiotemporal association rule base based on the spatiotemporal association rule base, and performing feature point combination processing to generate the key spatiotemporal feature set.
In this embodiment, the feature point extraction method refers to a technique of identifying representative spatio-temporal feature points from a dataset, which are typically data points that exhibit significant changes or patterns in a particular time and space. Common extraction methods include thresholding, edge detection, etc.
The preliminary feature point set is a set of initial feature points generated by a feature point extraction method. These points may contain redundant information that requires further processing to increase its representativeness.
The feature weighting method is a method for giving different weights to different feature points and is used for distinguishing the importance degree of each feature point. By calculating the contribution degree of each feature point, their roles in abnormal behavior can be reflected more accurately.
The weighted feature point set is a feature point set generated after feature weighting processing, wherein each feature point is attached with a corresponding weight value.
The association rule mining technique is a method for finding the internal relation between variables from a large amount of data, and can identify the space-time association between characteristic points and reveal hidden modes.
The spatio-temporal association rule base is a database storing various spatio-temporal association rules, each rule describing a relationship between different feature points.
The key space-time feature set is a key feature point set finally generated by carrying out combination processing on related space-time feature points in a space-time association rule base.
In the embodiment of the application, firstly, a system extracts representative space-time characteristic points through a characteristic point extraction method based on an abnormal behavior characteristic map to generate a preliminary characteristic point set, secondly, a characteristic weighting method is adopted to carry out weighting treatment on key space-time characteristic points in a dynamic capturing result, different weights are given according to the importance degrees of different characteristic points to generate a weighted characteristic point set, thirdly, an internal relation extraction treatment is carried out on space-time characteristics of different weighted characteristic points in the weighted characteristic point set through a relation rule mining technology to generate a space-time relation rule base, and finally, related space-time characteristic points in the space-time relation rule base are extracted based on the space-time relation rule base to carry out characteristic point combination treatment to generate the key space-time characteristic set.
The following is a specific example:
It is assumed that in an intelligent emergency transport pod, the system needs to accurately identify potential health risks. Firstly, a camera and a plurality of medical sensors are arranged in a cabin and used for monitoring the state of a patient in an unmanned emergency transportation process, data analysis is carried out through information such as facial expression, limb behavior and the like of the patient, secondly, a system extracts representative space-time characteristic points based on an abnormal physiological characteristic map through a characteristic point extraction method to generate a preliminary characteristic point set, such as heart rate fluctuation or respiratory rate change at a certain moment, thirdly, a characteristic weighting method is adopted to carry out weighting processing on the key space-time characteristic points, different weights are given according to importance degrees (such as heart rate fluctuation amplitude and respiratory jerkiness) of different characteristic points to generate a weighted characteristic point set, then, internal relation extraction processing is carried out on space-time characteristics of different weighted characteristic points in the weighted characteristic point set through a correlation rule mining technology to generate a space-time correlation rule base, potential relations among different characteristic points are revealed, and finally, based on the space-time correlation rule base, relevant space-time characteristic points such as severe respiratory characteristic areas frequently occurring in a certain time period are extracted, characteristic point combination processing is carried out to generate a key space-time characteristic set, and medical staff is helped to rapidly locate and respond to abnormal conditions.
In order to further improve accuracy of abnormal behavior identification and flexibility of response, in some embodiments, the abnormal behavior trend prediction model in step 103 employs a federal transition learning algorithm based on differential privacy, and the method comprises the steps of performing interval distribution analysis on risk scores output by the abnormal behavior trend prediction model based on the abnormal behavior trend prediction model, evaluating risk probabilities of different scoring intervals in the interval distribution analysis to generate a risk probability distribution map, performing emergency degree classification analysis on risk scores output by the abnormal behavior trend prediction model based on the risk probability distribution map by employing the federal transition learning algorithm based on differential privacy, extracting key features of the risk scores, dynamically setting classification thresholds of the risk scores to generate a preliminary emergency degree classification result, performing causal relation analysis on emergency degree classification results based on the preliminary emergency degree classification result, dynamically adjusting the classification thresholds to generate an optimized emergency degree classification result, and establishing a multi-stage feedback mechanism based on the optimized emergency degree classification result to set different feedback paths for different emergency degrees in the optimized emergency degree classification result to generate abnormal behavior causal relation.
In this embodiment, interval distribution analysis refers to the segmentation processing of risk scores, and the probability distribution condition of each scoring interval is evaluated, so that the risk levels of different scoring intervals can be better understood through the analysis, and a basis is provided for subsequent grading.
The risk probability distribution map is a visual tool, shows probability distribution conditions of different risk scoring intervals, and is helpful for intuitively identifying high-risk and low-risk areas and guiding emergency grading analysis.
The emergency level classification analysis is to divide abnormal behaviors into different emergency levels according to risk scores so as to take corresponding countermeasures.
The grading threshold is dynamically set as a grading standard for continuously adjusting the risk score according to actual conditions so as to adapt to changing environments and new threats.
Optimizing the emergency grading result is to dynamically adjust the grading threshold value by carrying out causal relation analysis on the preliminary emergency grading result, so as to generate a more scientific and reasonable final emergency grading result.
The multi-stage feedback mechanism is a multi-level feedback system, and different feedback paths are set for different levels of urgency.
In the embodiment of the application, firstly, a system performs interval distribution analysis on output risk scores based on an abnormal behavior trend prediction model, evaluates risk probabilities of different scoring intervals to generate a detailed risk probability distribution map, secondly, performs emergency degree classification analysis on the risk scores by utilizing a federal transition learning algorithm based on differential privacy, extracts key features and dynamically sets a classification threshold to generate a preliminary emergency degree classification result, thirdly, performs causal relation analysis on the preliminary emergency degree classification result by adopting a causal inference technology, dynamically adjusts the classification threshold to generate an optimized emergency degree classification result, and finally, establishes a multi-stage feedback mechanism based on the optimized emergency degree classification result, sets different feedback paths for emergency degrees of different stages to generate final abnormal behavior emergency degree.
The following is a specific example:
It is assumed that in an intelligent emergency delivery system, the system requires accurate assessment of potential health risks and timely response. Firstly, a camera and a plurality of medical sensors are arranged in a cabin and used for monitoring the state of a patient in an unmanned emergency transportation process, data analysis is carried out through information such as facial expressions, limb behaviors and the like of the patient, secondly, the system carries out interval distribution analysis on output risk scores based on an abnormal physiological trend prediction model, risk probabilities of different scoring intervals are evaluated, a detailed risk probability distribution map is generated, for example, which patients possibly have dyspnea or abnormal heart rate in a future period is displayed, thirdly, emergency classification analysis is carried out on the risk scores by utilizing a federal migration learning algorithm based on differential privacy, key characteristics such as heart rate fluctuation, respiratory frequency change and the like, classification thresholds are dynamically set, a preliminary emergency classification result is generated, then causal inference technology is adopted for carrying out relation analysis on the preliminary emergency classification result, for example, whether severe respiratory fluctuation in a specific time period leads to higher disease deterioration risk is explored, the emergency classification threshold is dynamically adjusted, an optimized emergency classification result is generated, finally, a multi-stage feedback mechanism is established based on the optimized emergency classification result, different emergency feedback paths are set for different emergency grades, such as automatic follow-up alarms or automatic response alarms are sent to medical personnel, the emergency response is improved, and the emergency response is finally, the comfort of the emergency response is ensured, and the ventilation is improved, and the comfort of the medical device is improved.
In order to further improve accuracy of abnormal behavior identification and flexibility of response, in some embodiments, the step 103 of applying a federal transition learning algorithm based on differential privacy to the risk probability distribution diagram includes deep parsing based on the risk probability distribution diagram, focusing on overall risk patterns and trends in the risk probability distribution diagram, vectorizing the overall risk patterns and trends to generate a risk feature matrix, applying a federal transition learning algorithm based on differential privacy to each risk feature vector in the risk feature matrix based on the risk feature matrix, performing emergency grading analysis on risk scores output by the abnormal behavior trend prediction model to generate emergency grading scores, performing score fusion processing based on key features of the risk scores, dynamically setting grading thresholds of the risk scores according to score fusion processing results, generating an optimal grading threshold, introducing an adaptive adjustment mechanism based on the optimal grading thresholds, recording output data in real time during running of the abnormal behavior prediction model, and performing contrast analysis with the optimal grading thresholds to automatically adjust, and generating a preliminary emergency grading result.
In this embodiment, the risk probability distribution map is a visualization tool, showing probability distribution of different risk score intervals.
The overall risk pattern and trend refer to global risk features reflected in the risk probability distribution map, including the trend of variation, peaks and valleys of the risk level, and the like.
The risk feature matrix is a matrix composed of a plurality of risk feature vectors, each representing a risk feature at a particular point in time or spatial location.
The emergency grading score is a result of grading the abnormal behavior risk score and reflects the emergency degree of different events. The higher the score, the more urgent the event, requiring priority handling.
The scoring fusion process comprehensively processes the results of the plurality of scoring methods to generate more comprehensive and accurate scoring results.
The optimized grading threshold is a grading standard for continuously adjusting the risk score according to actual conditions so as to adapt to changing environments and new threats.
The self-adaptive adjustment mechanism is a mechanism capable of automatically adjusting system parameters according to real-time data, and can continuously optimize the performance of the system during operation, so as to ensure that the system is always in an optimal state.
In the embodiment of the application, firstly, a system carries out deep analysis based on a risk probability distribution diagram, focuses on an overall risk mode and trend, carries out vectorization processing on the overall risk mode and trend to generate a risk feature matrix, secondly, carries out emergency grading analysis on each risk feature vector in the risk feature matrix by utilizing a federal migration learning algorithm based on differential privacy to generate emergency grading, thirdly, carries out grading fusion processing on key features of the emergency grading, dynamically sets a grading threshold of the risk grading according to a fusion processing result to generate an optimized grading threshold, and finally, introduces an adaptive adjustment mechanism, records output data in real time during operation of an abnormal behavior trend prediction model, and carries out comparison analysis with the optimized grading threshold to automatically adjust to generate a preliminary emergency grading result.
The following is a specific example:
the method comprises the steps of providing an intelligent emergency rescue cabin, enabling a system to accurately evaluate and timely respond to potential health risks, firstly generating a risk feature matrix based on data collected from a camera and a medical sensor, focusing on an overall health mode and trend, for example, the change trend of vital signs of a patient in a certain period of time, carrying out vectorization processing on the risk feature matrix to provide a basis for subsequent analysis, secondly carrying out emergency grading analysis on each risk feature vector in the risk feature matrix by utilizing a federal migration learning algorithm based on differential privacy, for example, evaluating health risk grades in different time periods to generate emergency grading scores to ensure accurate evaluation of patient conditions, thirdly, carrying out grading fusion processing on key features (such as heart rate, respiratory rate and blood oxygen saturation) of the emergency grading scores, dynamically setting grading thresholds of the risk grading scores according to fusion processing results to generate optimal grading thresholds so as to reflect the current health state of the patient more accurately, and finally introducing a self-adaptive adjustment mechanism to record output data in real time during operation of an abnormal physiological trend prediction model, for example, carrying out emergency grading analysis with the emergency grading thresholds to compare with the emergency grading thresholds to generate the emergency grading scores to ensure accurate evaluation results. This ensures that the system is able to respond quickly and effectively to sudden health changes in the patient, such as when manic symptoms are detected in the patient, and the corresponding sedation procedure may be initiated immediately.
In order to further improve response efficiency of abnormal behavior identification and intellectualization of an early warning strategy, in some embodiments, the step 104 of outputting a customized safety notification based on the emergency degree of abnormal behavior includes outputting a customized safety notification based on the emergency degree of abnormal behavior and sending the customized safety notification to a target source group of the real-time environment-aware data to generate a customized safety notification record, setting an automatic response flow based on the customized safety notification record to automate the response flow of the target source group after receiving the customized safety notification record to generate an automatic response log, continuously monitoring user behavior and environment change of the target source group based on the automatic response log to capture potential abnormal activities in the continuous monitoring process to generate a real-time monitoring report, extracting a newly generated abnormal mode in the real-time monitoring report based on the real-time monitoring report to dynamically update the abnormal mode library to generate the intelligent identification and early warning strategy.
In this embodiment, the customized security notification is based on the degree of urgency of the abnormal behavior, and the system will generate security pre-warning information for a particular group or scenario.
The customized security notification record refers to notification details, including a receiving object, notification time, content abstract, and the like, automatically recorded by the system after the customized security notification is generated and sent.
The automatic response flow refers to a preset series of automatic operations, and the goal source group can be ensured to quickly take corresponding measures after receiving the customized security notification.
The automatic response log records the execution condition of the automatic response flow, including information such as response time, execution result and the like.
The real-time monitoring report is a report generated by a continuous monitoring process, and details the user behavior and environment change of a target source group and potential abnormal activities captured in the process are recorded.
The newly generated abnormal pattern refers to a novel abnormal behavior pattern found in the real-time monitoring report that does not appear in the abnormal pattern library.
The intelligent recognition and early warning strategy is to dynamically update an abnormal mode library according to the real-time monitoring report and the new abnormal mode, optimize the recognition capability and early warning mechanism of the system, and ensure that the system can adapt to new threats and changes and keep long-term effectiveness.
The method comprises the steps of firstly outputting customized safety notification based on emergency degree of abnormal behaviors by a system, sending the customized safety notification to a target source group of real-time environment perception data, generating customized safety notification records at the same time, secondly setting an automatic response flow based on the customized safety notification records, ensuring that the response flow of the target source group after receiving the notification is automatic, generating an automatic response log, continuously monitoring user behaviors and environment changes of the target source group based on the automatic response log, capturing potential abnormal activities in a continuous monitoring process, generating a real-time monitoring report, and finally extracting new abnormal modes based on the real-time monitoring report to dynamically update an abnormal mode library to generate an intelligent recognition and early warning strategy.
The following is a specific example:
It is assumed that in an intelligent multifunctional emergency transport pod, the system needs to improve response speed and early warning accuracy to patient health risks. Firstly, a camera and a plurality of medical sensors are arranged in a cabin and used for monitoring the state of a patient in an unmanned emergency treatment process, data analysis is carried out by analyzing information such as facial expression, limb behavior and the like of the patient, the adjustment of breathing machine parameters is assisted to guide to enhance comfortableness and optimize respiratory support (such as oxygen concentration improvement, tidal volume adjustment and minute ventilation) and the adjustment of a syringe pump is carried out to carry out emergency treatment such as sedative bolus injection or epinephrine bolus injection when necessary, secondly, a customized safety notification is output based on abnormal physiological emergency and sent to a follow-up medical staff and a remote medical center, and meanwhile, a customized safety notification record is generated, such as early warning of serious illness state of a certain patient is generated, and thirdly, an automatic response flow is set based on the customized safety notification record, such as automatic adjustment of medical equipment parameters in the cabin, starting of an emergency treatment program and automatic response log generation, finally, the continuous monitoring is carried out on the behavior change and vital sign of the patient based on the automatic response log, a real-time monitoring report is generated, the system is ensured to be capable of timely adjusting a treatment scheme and dynamically updating an abnormal mode library, the intelligent recognition strategy is generated, and the system is ensured to be capable of better corresponding to future conditions.
The present application considers that, in order to solve the problems of insufficient recognition accuracy of abnormal behaviors, untimely response and limited processing capacity on complex space-time correlation in the prior art, the embodiment of the application proposes the alternative to solve the problems, so a new alternative is proposed, which includes:
based on the preprocessing data set, a space-time dynamic graph nerve flow algorithm is applied, and the time sequence data of abnormal activities in the abnormal activity list and the spatial relevance are dynamically captured through a continuous time graph rolling mechanism in the space-time dynamic graph nerve flow algorithm, so that a dynamic capturing result is generated, and the method comprises the following steps:
Based on the preprocessing data set, short-term fluctuation is eliminated through a time sequence smoothing technology, space coordinates of different scales are unified to the same level by using a space coordinate normalization method, and abnormal activity intensity normalization processing is adopted to generate a preliminary characteristic response value;
The preliminary feature response value is calculated by the following formula:
R(t)=η·exp(-λ·|T(t)-T0|-v·|S(t)-S0|)+k·|A(t)-A0|
Wherein R (T) is a preliminary characteristic response value of time series data of abnormal activity at time T and space correlation, eta is a response intensity coefficient, lambda is a time deviation attenuation factor, T (T) and T0 are time stamps at time T and reference time respectively, v is a position deviation correction coefficient, S (T) and S0 are space coordinates at time T and reference time respectively, kappa is an activity intensity weight, A (T) and A0 are abnormal activity intensities at time T and reference time respectively, and T is a time variable used for referring to specific time;
Based on the preliminary characteristic response values, identifying node groups with similar characteristic response modes by utilizing characteristic value cluster analysis, enhancing the degree of distinction between characteristics by nonlinear transformation, capturing potential periodic abnormal behavior modes by combining time and space periodic analysis, and introducing a weighted comprehensive evaluation mechanism to generate a comprehensive dynamic capture index;
The integrated dynamic capture index is calculated by the following formula:
The method comprises the steps of (1) obtaining a comprehensive dynamic capture index of a moment T, wherein i is a node index which participates in calculation, from 1 to n, n is the number of nodes which participate in calculation, alphai is a response adjustment coefficient of an ith node, betai is a nonlinear response adjustment factor of the ith node, R (T) is a preliminary characteristic response value of the moment T, gammai is a time period coefficient of the ith node, deltai is a time frequency factor of the ith node, T (T) and T0 are respectively time segments of the moment T and a reference moment, thetai is a position period coefficient of the ith node, Ei is a position frequency factor of the ith node, and S (T) and S0 are respectively spatial coordinates of the moment T and the reference moment;
based on the comprehensive dynamic capture index, identifying abnormal events which deviate from a normal range remarkably through threshold screening and abnormal point detection, classifying and labeling, estimating future abnormal behavior trend by applying a time sequence prediction model, providing early warning information and generating a dynamic capture result.
The method aims to improve the accuracy and response efficiency of abnormal behavior identification, quantizes the influence of time sequence data and space relativity of abnormal activities through a preliminary characteristic response value, further enhances the distinguishing degree between characteristics through nonlinear transformation and periodical analysis on the basis of the preliminary characteristic response value through comprehensive dynamic capture indexes, improves the identification capacity and early warning accuracy of a system, solves the problems in the prior art, and provides a more scientific and effective solution for the intelligent identification and early warning field.
It is assumed that in an intelligent emergency medical pod, the system needs to monitor the health status of the patient in real time and pre-warn of potential health deterioration conditions in advance. Let time stamp T (T) = 1678456789, reference time T0 = 1678456780, spatial coordinate S (T) = (40.7128, -74.0060), reference coordinate S0 = (40.7128, -74.0059), abnormal activity intensity a (T) =0.85, reference intensity a0 =0.5, coefficient η=1.0, λ=0.01, v=0.02, κ=0.5;
R(t)=
1.0·exp(-0.01·|1678456789-1678456780|-0.02·|40.7128-40.7128|)+0.5·∣
0.85-0.5∣=0.775;
let n=3 nodes involved in the calculation, let a1＝0.8,β1＝0.05,γ1＝0.6,δ1＝0.01,θ1＝0.4,∈1 =0.02 for node 1;
Assuming a set threshold of 2, since the calculated result 2.15 is greater than the set threshold, it is indicated that serious health deterioration of the patient may occur in a future period of time. This is because the higher integrated dynamic capture index reflects the high sensitivity of the system to changes in the patient's condition, while also verifying the effectiveness of abnormal behavior identification and early warning. Through the steps, the real-time monitoring and early warning of the state of the patient are ensured, the efficiency and scientificity of emergency treatment are improved, the response speed and reliability of the whole system are enhanced, and the treatment scheme is timely adjusted, so that the emergency effect and the safety of the patient are improved. The application considers that, in order to solve the problems of insufficient recognition precision of abnormal behaviors, inaccurate risk assessment and limited processing capacity on complex space-time correlation in the prior art, the embodiment of the application proposes the alternative scheme to solve the problems. A new alternative is therefore proposed, which comprises:
Based on the risk feature matrix, applying a federal transition learning algorithm based on differential privacy to each risk feature vector in the risk feature matrix, performing emergency grading analysis on risk scores output by the abnormal behavior trend prediction model, and generating emergency grading scores, wherein the emergency grading scores comprise:
based on a risk feature matrix, carrying out standardization on risk features of different scales in the risk feature matrix through standardization processing, extracting main feature vectors in the standardization result by applying principal component analysis, reducing data dimension and retaining most remarkable information, and identifying abnormal values in the main feature vectors by adopting an abnormal value detection technology so as to generate a risk feature weighting index;
the risk feature weighting index is calculated by the following formula:
Wherein Wj is the risk feature weighting index of the jth risk feature vector, ζ is the comprehensive adjustment coefficient, Xp is the time sequence weight factor of the jth feature, Rjp and R0p are the time sequence data of the jth and reference risk feature vectors at the jth feature respectively, ψ is the feature deviation correction coefficient, φ is the feature influence factor, Fjp and F0p are the spatial correlation features of the jth and reference risk feature vectors at the jth feature respectively, p is the index of the feature, from 1 to m, and m is the feature quantity;
Based on the risk characteristic weighting indexes, classifying nodes with similar characteristic weighting indexes in the risk characteristic weighting indexes into a group by utilizing cluster analysis, amplifying differences among different node groups by nonlinear transformation, and introducing a time sequence prediction model to capture time-varying trends of the differences among the node groups so as to generate a risk response value;
the risk response value is calculated by the following formula:
Wherein Hi is the risk response value of the ith risk feature vector, r is the index of the feature dimension involved in calculation from 1 to l, l is the number of feature dimensions involved in calculation, alphar is the response adjustment coefficient of the ith feature dimension, lambdar is the nonlinear adjustment factor of the ith feature dimension, Wi is the risk feature weighting index of the ith risk feature vector, betar is the time sequence difference weight of the ith feature dimension, deltar is the time frequency factor of the ith feature dimension, Cir and C0r are the time sequence features of the ith and reference risk feature vectors in the ith feature dimension respectively, thetar is the position cycle coefficient of the ith feature dimension, Er is the position frequency factor of the ith feature dimension, Tir and T0r are the time stamps of the ith and reference risk feature vectors in the ith feature dimension respectively, rhor is the space influence coefficient of the ith feature dimension, gammar is the space sensitivity factor of the ith feature dimension, Sir and S0r are the space coordinates of the ith feature vector in the ith feature dimension respectively;
And performing further threshold screening and classification labeling based on the risk response value, identifying high-risk nodes in the threshold screening and classification labeling result to perform emergency grading analysis, and introducing an adaptive threshold adjustment mechanism to generate emergency grading scores.
The method aims at improving the accuracy of abnormal behavior identification and the effectiveness of risk assessment, reducing the data dimension through quantification of the importance of risk feature vectors and through standardization processing and principal component analysis, retaining the most remarkable information, amplifying the difference between different node groups on the basis of risk feature weighting indexes through clustering analysis and nonlinear transformation, and introducing a time sequence prediction model to capture time-varying trend, so that a more accurate risk assessment result is generated.
In the risk feature weighting index, time series distance itemsThe partial term measures the distance of the risk feature vector on the time sequence, enhances the sensitivity to small distance changes by the form of the inverse square root, and is a spatial correlation deviation termThe partial terms reflect the deviation of the spatial correlation characteristics, and the tiny difference between the characteristics is amplified through a logarithmic function and a product form;
Wherein Rjp and R0p are respectively time series data of the jth risk feature vector and the reference risk feature vector in the p-th feature, and can be obtained through historical data or real-time monitoring, Fjp and F0p are respectively spatial correlation features of the jth risk feature vector and the reference risk feature vector in the p-th feature, and can be obtained through spatial coordinates or other position information, and χp, ζ, ψ and φ are preset coefficients, and are usually determined according to historical data statistical analysis or expert experience;
In the risk response value, a nonlinear response adjustment termThe subitem enhances the influence of the risk characteristic weighting index through a hyperbolic sine function and an exponential decay function, considers the difference of time sequence characteristics, and a time period item thetar·arctan(∈r·(Tir-T0r), wherein the subitem captures the periodic change in time, reflects the difference between timestamps through an arctangent function, and affects the items in spaceThe sub-term reflects the difference between the space coordinates, and the influence of the space distance is amplified in a form of a fourth power;
the method comprises the steps of calculating Wi by a risk feature weighted index formula, obtaining time sequence features of an ith risk feature vector and a reference risk feature vector in an (r) feature dimension through historical data or real-time monitoring, obtaining time stamps of the ith risk feature vector and the reference risk feature vector in the (r) feature dimension through a system built-in clock or external time source synchronization, obtaining space coordinates of the ith risk feature vector and the reference risk feature vector in the (r) feature dimension through a GPS or other positioning equipment, and setting alphar,λr,βr,δr,θr,∈r,ρr and gammar to parameters set for each feature dimension through machine learning algorithm training or expert experience, wherein Cir and C0r are respectively;
It is assumed that in an intelligent emergency medical cabin, the system needs to monitor potential health risks in real time and early warn in advance, time series data Rjp = [0.85,0.9], reference time series data R0p = [0.6,0.7], spatial correlation feature Fjp = [0.75,0.7], reference spatial correlation feature F0p = [0.55,0.6], comprehensive adjustment coefficient ζ=1.2, time series weight factor χp = [0.9,0.8], feature deviation correction coefficient ψ=0.6, and feature influence factor φ=0.3;
let the number of feature dimensions involved in the calculation l=3, set for the 1 st feature dimension α1＝0.7,λ1＝0.04,β1＝0.02,δ1＝0.03,θ1＝0.5,∈1＝0.04,ρ1＝0.6,γ1＝0.05;
Assuming a set threshold of 1.1, since the calculated result 1.32 is greater than the set threshold, it is indicated that the patient has a higher health risk, reflecting the high sensitivity of the system to the patient's condition, and verifying the effectiveness of the early warning mechanism. Through the steps, the real-time monitoring and early warning of potential health risks are ensured, the emergency treatment efficiency and scientificity of the emergency medical cabin are improved, the response speed and reliability of the whole system are enhanced, medical teams can intervene in treatment earlier, and the survival probability and rehabilitation quality of patients are improved. The intelligent solution not only can quickly respond to sudden health changes of patients, but also helps to optimize medical resource allocation and ensure optimal treatment effect. Fig. 2 is a schematic structural diagram of an intelligent recognition and early warning system according to an embodiment of the present application, where, as shown in fig. 2, the device includes:
The receiving module 21 is configured to receive real-time environment sensing data and predefine an abnormal pattern library, and perform dynamic matching processing on the real-time environment sensing data based on the abnormal pattern library to generate a potential abnormal activity list;
The processing module 22 is configured to dynamically capture time sequence data and spatial correlation of abnormal activities in the abnormal activity list by using a continuous time graph convolution mechanism in the spatio-temporal dynamic graph neural flow algorithm based on the potential abnormal activity list and applying a spatio-temporal dynamic graph neural flow algorithm, and perform weighting processing on key spatio-temporal feature points in a dynamic capture result by using a feature weighting method to generate an abnormal behavior trend prediction model;
The analysis module 23 is configured to perform emergency degree classification analysis on risk scores output by the abnormal behavior trend prediction model by using a federal transition learning algorithm based on differential privacy, extract key features of the risk scores, dynamically set classification thresholds of the risk scores, perform causal relationship analysis on emergency degree classification analysis results by using a causal inference technology, and dynamically adjust the classification thresholds to generate abnormal behavior emergency degrees;
The monitoring module 24 is configured to output a customized security notification based on the emergency degree of the abnormal behavior, send the customized security notification to a target source group of the real-time environment-aware data, continuously monitor a user behavior and an environment change of the target source group, dynamically update the abnormal pattern library, and generate an intelligent recognition and early warning policy.
An intelligent recognition and early warning system shown in fig. 2 may execute an intelligent recognition and early warning method shown in the embodiment shown in fig. 1, and its implementation principle and technical effects are not repeated. The specific manner in which the various modules and units of the intelligent recognition and early warning system perform the operations in the foregoing embodiments has been described in detail in the embodiments related to the method, and will not be described in detail herein.
In one possible design, an intelligent recognition and early warning system of the embodiment of FIG. 2 may be implemented as a computing device, as shown in FIG. 3, that may include a storage component 31 and a processing component 32;
the storage component 31 stores one or more computer instructions for execution by the processing component 32.
The processing component 32 is configured to receive real-time environment sensing data and predefine an abnormal pattern library, dynamically match the real-time environment sensing data based on the abnormal pattern library to generate a potential abnormal activity list, dynamically set a classification threshold of the risk score based on the potential abnormal activity list, dynamically adjust the classification threshold by a continuous time graph convolution mechanism in the spatio-temporal dynamic graph nerve flow algorithm based on the potential abnormal activity list, dynamically capture time series data and spatial correlation of abnormal activities in the abnormal activity list, weighting key spatio-temporal feature points in a dynamic capture result by a feature weighting method to generate an abnormal behavior trend prediction model, dynamically analyze the risk score output by the abnormal behavior trend prediction model by a federal migration learning algorithm based on difference privacy, dynamically set the classification threshold of the risk score based on the potential abnormal activity list, dynamically adjust the classification threshold by a causal relationship analysis result by an inference technology, generate abnormal behavior, customize a safety notification based on the abnormal behavior emergency perception, send the abnormal behavior trend prediction model to a target source, and dynamically identify a new user population, and dynamically identify a new source of the abnormal behavior, and a new user population.
Wherein the processing component 32 may include one or more processors to execute computer instructions to perform all or part of the steps of the methods described above. Of course, the processing component may also be implemented as one or more Application Specific Integrated Circuits (ASICs), digital Signal Processors (DSPs), digital Signal Processing Devices (DSPDs), programmable Logic Devices (PLDs), field Programmable Gate Arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic elements for executing the methods described above.
The storage component 31 is configured to store various types of data to support operations at the terminal. The memory component may be implemented by any type or combination of volatile or nonvolatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disk.
Of course, the computing device may necessarily include other components as well, such as input/output interfaces, display components, communication components, and the like.
The input/output interface provides an interface between the processing component and a peripheral interface module, which may be an output device, an input device, etc.
The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, and the like.
The computing device may be a physical device or an elastic computing host provided by the cloud computing platform, and at this time, the computing device may be a cloud server, and the processing component, the storage component, and the like may be a base server resource rented or purchased from the cloud computing platform.
The embodiment of the application also provides a computer storage medium which stores a computer program, and the computer program can realize the intelligent identification and early warning method of the embodiment shown in the figure 1 when being executed by a computer.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
It should be noted that the above-mentioned embodiments are merely for illustrating the technical solution of the present application, and not for limiting the same, and although the present application has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solution described in the above-mentioned embodiments may be modified or some technical features may be equivalently replaced, and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the spirit and scope of the technical solution of the embodiments of the present application.