Intelligent identification and early warning method and system

Abstract

The invention provides an intelligent identification and early warning method and system. The method comprises the following steps: receiving real-time environment sensing data, performing dynamic matching processing, and generating a potential abnormal activity list; on the basis of the potential abnormal activity list, dynamically capturing time sequence data and spatial relevance by applying a spatio-temporal dynamic graph neural flow algorithm and through a continuous time graph convolution mechanism, and weighting key spatio-temporal feature points by adopting a feature weighting method to generate an abnormal behavior trend prediction model; based on the abnormal behavior trend prediction model, carrying out emergency degree grading analysis by using a federated transfer learning algorithm based on differential privacy, and carrying out causal relationship analysis by using a causal inference technology to generate an abnormal behavior emergency degree; and outputting a customized safety notification based on the abnormal behavior emergency degree, sending the customized safety notification to a target source group, performing continuous monitoring, and generating an intelligent identification and early warning strategy. According to the technical scheme provided by the invention, the intelligent identification accuracy and the early warning response timeliness are remarkably improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of intelligent security monitoring technology, and in particular to an intelligent identification and early warning method and system. Background Art

[0002] With the rapid development of smart cities and IoT technologies, the application of real-time environmental perception data is becoming more and more widespread. In smart cities, smart communities, rail transit, industrial Internet and other fields, the demand for intelligent identification and early warning of abnormal behaviors is growing. These application scenarios require the system to quickly and accurately capture potential abnormal activities and issue early warning notifications in a timely manner to ensure public safety and improve management efficiency.

[0003] At present, there are many abnormal behavior recognition and early warning systems on the market, mainly based on traditional machine learning algorithms and rule engines. These systems analyze the collected environmental perception data through predefined rules or models, generate a preliminary list of abnormal activities, and issue early warning notifications based on established thresholds; in addition, some advanced systems have begun to introduce deep learning algorithms, such as convolutional neural networks and recurrent neural networks, to improve the accuracy of anomaly detection.

[0004] Although existing solutions have met some of the needs to a certain extent, there are still significant shortcomings; traditional methods rely on predefined rules and static models, which are difficult to cope with complex and changeable real-world environments, resulting in low accuracy in identifying abnormal behaviors and insufficient early warning responses, resulting in insufficient accuracy in intelligent identification and timeliness in early warning responses; most existing systems fail to make full use of time series data and spatial correlations, and are unable to fully capture the dynamic changes in abnormal activities, affecting the reliability of predictions; in the process of data analysis, user privacy protection measures are insufficient, which easily leads to data leakage risks and reduces user trust. Summary of the invention

[0005] The embodiments of the present application provide an intelligent identification and early warning method and system to solve the problems of insufficient accuracy of intelligent identification and timeliness of early warning response in the prior art.

[0006] In a first aspect, an embodiment of the present application provides an intelligent identification and early warning method, including:

[0007] Receive real-time environment perception data and predefine an abnormal pattern library, and based on the abnormal pattern library, dynamically match the real-time environment perception data to generate a potential abnormal activity list;

[0008] Based on the potential abnormal activity list, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and the key spatiotemporal feature points in the dynamic capture results are weighted by a feature weighting method to generate an abnormal behavior trend prediction model;

[0009] Based on the abnormal behavior trend prediction model, a federated transfer learning algorithm based on differential privacy is used to perform an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate the abnormal behavior urgency;

[0010] Based on the urgency of the abnormal behavior, a customized safety notification is output and sent to the target source group of the real-time environmental perception data, the user behavior and environmental changes of the target source group are continuously monitored, the abnormal pattern library is dynamically updated, and an intelligent recognition and early warning strategy is generated.

[0011] Optionally, based on the potential abnormal activity list, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and a feature weighting method is used to weight key spatiotemporal feature points in the dynamic capture results to generate an abnormal behavior trend prediction model, including:

[0012] Based on the potential abnormal activity list, using multi-source information fusion technology, collecting heterogeneous data in the potential abnormal activity list for integration and synchronization processing to generate a heterogeneous data set;

[0013] Based on the heterogeneous data set, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm to generate an abnormal behavior feature map;

[0014] Based on the abnormal behavior feature map, a feature weighting method is used to perform weighted processing on key spatiotemporal feature points in the dynamic capture results, and feature points with important influence in the weighted processing results are identified to generate a key spatiotemporal feature set;

[0015] Based on the key spatiotemporal feature set and combined with the modeling capability of the spatiotemporal dynamic graph neural flow algorithm, the time series characteristics of the abnormal behavior in the abnormal behavior feature map are modeled to generate an abnormal behavior trend prediction model.

[0016] Optionally, based on the heterogeneous data set, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm to generate an abnormal behavior feature map, including:

[0017] Based on the heterogeneous data set, using data cleaning and denoising technology, extracting time series data and spatial correlation information in the heterogeneous data set for preprocessing to generate a preprocessed data set;

[0018] Based on the preprocessed data set, using the spatiotemporal dynamic graph neural flow algorithm, through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, the time series data and spatial correlation of the abnormal activities in the abnormal activity list are dynamically captured to generate a dynamic capture result;

[0019] Based on the dynamic capture results, applying pattern recognition technology, classifying and clustering the abnormal patterns in the dynamic capture results to generate an abnormal pattern library;

[0020] Based on the abnormal pattern library, a multi-dimensional icon is constructed to perform a visual analysis of the abnormal behaviors in the abnormal pattern library, display the spatiotemporal distribution characteristics of the abnormal behaviors, and generate an abnormal behavior feature map.

[0021] Optionally, based on the abnormal behavior feature map, a feature weighting method is used to perform weighted processing on key spatiotemporal feature points in the dynamic capture result, and feature points with important influence in the weighted processing result are identified to generate a key spatiotemporal feature set, including:

[0022] Based on the abnormal behavior feature map, extracting and processing representative spatiotemporal feature points in the abnormal behavior feature map by a feature point extraction method to generate a preliminary feature point set;

[0023] Based on the preliminary feature point set, a feature weighting method is used to perform weighted processing on the key spatiotemporal feature points in the dynamic capture result, and different weights are assigned to different feature points in the key spatiotemporal feature points to distinguish the importance of the different feature points, thereby generating a weighted feature point set;

[0024] Based on the weighted feature point set, the temporal and spatial features of different weighted feature points in the weighted feature point set are extracted and processed by association rule mining technology to generate a temporal and spatial association rule library;

[0025] Based on the spatiotemporal association rule base, relevant spatiotemporal feature points in the spatiotemporal association rule base are extracted, feature point combination processing is performed, and a key spatiotemporal feature set is generated.

[0026] Optionally, based on the abnormal behavior trend prediction model, a federated transfer learning algorithm based on differential privacy is used to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis result, dynamically adjust the grading threshold, and generate abnormal behavior urgency, including:

[0027] Based on the abnormal behavior trend prediction model, performing interval distribution analysis on the risk score output by the abnormal behavior trend prediction model, evaluating the risk probability of different score intervals in the interval distribution analysis, and generating a risk probability distribution graph;

[0028] Based on the risk probability distribution graph, a federated transfer learning algorithm based on differential privacy is used to perform an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, and generate a preliminary urgency grading result;

[0029] Based on the preliminary urgency grading result, using causal inference technology, performing causal relationship analysis on the urgency grading analysis result, dynamically adjusting the grading threshold, and generating an optimized urgency grading result;

[0030] Based on the optimized urgency grading result, a multi-level feedback mechanism is established, and different feedback paths are set for different levels of urgency in the optimized urgency grading result to generate abnormal behavior urgency.

[0031] Optionally, based on the risk probability distribution graph, a federated transfer learning algorithm based on differential privacy is used to perform an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set a grading threshold for the risk score, and generate a preliminary urgency grading result, including:

[0032] Performing in-depth analysis based on the risk probability distribution graph, focusing on the overall risk pattern and trend in the risk probability distribution graph, vectorizing the overall risk pattern and trend, and generating a risk feature matrix;

[0033] Based on the risk feature matrix, applying a federated transfer learning algorithm based on differential privacy to each risk feature vector in the risk feature matrix, performing an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, and generating an urgency grading score;

[0034] Based on the urgency grading score, the key features of the risk score are extracted to perform score fusion processing, and according to the score fusion processing result, the grading threshold of the risk score is dynamically set to generate an optimized grading threshold;

[0035] Based on the optimized classification threshold, an adaptive adjustment mechanism is introduced to record output data in real time during the operation of the abnormal behavior trend prediction model, and compare and analyze the optimized classification threshold to automatically adjust and generate a preliminary urgency classification result.

[0036] Optionally, the customized safety notification is output based on the urgency of the abnormal behavior and sent to the target source group of the real-time environmental perception data, the user behavior and environmental changes of the target source group are continuously monitored, the abnormal pattern library is dynamically updated, and an intelligent identification and early warning strategy is generated, including:

[0037] Outputting a customized safety notification based on the urgency of the abnormal behavior, and sending it to a target source group of the real-time environmental perception data, and generating a customized safety notification record;

[0038] Based on the customized security notification record, an automated response process is set to automate the response process of the target source group after receiving the customized security notification record, and an automated response log is generated;

[0039] Based on the automated response log, continuously monitor the user behavior and environmental changes of the target source group, capture potential abnormal activities during the continuous monitoring process, and generate a real-time monitoring report;

[0040] Based on the real-time monitoring report, newly generated abnormal patterns in the real-time monitoring report are extracted to dynamically update the abnormal pattern library and generate intelligent identification and early warning strategies.

[0041] In a second aspect, an embodiment of the present application provides an intelligent identification and early warning system, including:

[0042] A receiving module, used for receiving real-time environment perception data and predefining an abnormal pattern library, and performing dynamic matching processing on the real-time environment perception data based on the abnormal pattern library to generate a potential abnormal activity list;

[0043] A processing module is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list based on the potential abnormal activity list by using a spatiotemporal dynamic graph neural flow algorithm, and to perform weighted processing on key spatiotemporal feature points in the dynamic capture results by using a feature weighting method to generate an abnormal behavior trend prediction model;

[0044] An analysis module is used to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model based on the abnormal behavior trend prediction model by using a federated transfer learning algorithm based on differential privacy, extract key features of the risk score, dynamically set a grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate abnormal behavior urgency;

[0045] The monitoring module is used to output customized safety notifications based on the urgency of the abnormal behavior, send them to the target source group of the real-time environmental perception data, continuously monitor the user behavior and environmental changes of the target source group, dynamically update the abnormal pattern library, and generate intelligent recognition and early warning strategies.

[0046] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an intelligent identification and early warning method as described in the first aspect.

[0047] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an intelligent identification and early warning method as described in the first aspect.

[0048] In the embodiment of the present application, real-time environmental perception data is received and an abnormal pattern library is predefined. Based on the abnormal pattern library, the real-time environmental perception data is dynamically matched and processed to generate a potential abnormal activity list. Based on the potential abnormal activity list, a spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm. A feature weighting method is used to weight key spatiotemporal feature points in the dynamic capture results to generate an abnormal behavior trend prediction model. Based on the abnormal behavior trend prediction model, the algorithm is used to perform a weighted processing on the key spatiotemporal feature points in the dynamic capture results to generate an abnormal behavior trend prediction model. Using a federated transfer learning algorithm based on differential privacy, the risk score output by the abnormal behavior trend prediction model is analyzed for urgency grading, the key features of the risk score are extracted, the grading threshold of the risk score is dynamically set, and the causal inference technology is used to analyze the causal relationship of the urgency grading analysis results, and the grading threshold is dynamically adjusted to generate the urgency of abnormal behavior; based on the urgency of abnormal behavior, a customized security notification is output and sent to the target source group of the real-time environmental perception data, the user behavior and environmental changes of the target source group are continuously monitored, the abnormal pattern library is dynamically updated, and an intelligent recognition and early warning strategy is generated. Through dynamic matching processing of real-time environmental perception data and predefined abnormal pattern library, the accurate capture of potential abnormal activities is ensured; the spatiotemporal dynamic graph neural flow algorithm is adopted, which can effectively handle complex spatiotemporal correlations and improve the adaptability to different environmental changes; based on the urgency of abnormal behavior, a customized security notification is output to achieve accurate early warning for specific groups and improve the response speed and effectiveness; by continuously monitoring the behavior and environmental changes of the target source group and dynamically updating the abnormal pattern library, the adaptability and long-term reliability of the system are ensured; differential privacy technology is introduced in the analysis process to ensure the security and privacy of user data.

[0049] Furthermore, heterogeneous data are integrated through multi-source information fusion technology to generate comprehensive and consistent heterogeneous data sets, which enhances the availability and accuracy of the data; the feature weighting method is used to identify key spatiotemporal feature points, highlighting the influence of important features and improving the explanatory power and prediction accuracy of the model; combined with the modeling capabilities of the spatiotemporal dynamic graph neural flow algorithm, an abnormal behavior trend prediction model is generated to provide a scientific basis for decision-making and enhance the reliability and practicality of the prediction.

[0050] Furthermore, by performing interval distribution analysis on risk scores and evaluating the risk probabilities in different score intervals, an accurate risk probability distribution map is generated, providing a solid foundation for subsequent analysis; a federated transfer learning algorithm based on differential privacy is used to perform urgency grading analysis, and the threshold is dynamically adjusted in combination with causal inference technology to ensure the scientificity and rationality of the grading results; a multi-level feedback mechanism is established, and different feedback paths are set for different levels of urgency, which enhances the flexibility and response speed of the system and ensures the effectiveness of the early warning; differential privacy technology is applied throughout the process to ensure that user privacy is not leaked during data analysis, thereby enhancing user trust.

[0051] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A flowchart of an intelligent identification and early warning method provided in an embodiment of the present application;

[0054] Figure 2 A schematic diagram of the structure of an intelligent identification and early warning system provided in an embodiment of the present application;

[0055] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0057] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0059] Figure 1 A flowchart of an intelligent identification and early warning method is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0060] 101. Receive real-time environment perception data and predefine an abnormal pattern library, and based on the abnormal pattern library, dynamically match the real-time environment perception data to generate a potential abnormal activity list;

[0061] In this step, real-time environmental perception data refers to environmental information collected by sensors, cameras and other devices within a specific time period, such as temperature, humidity, light intensity, object movement trajectory, etc.

[0062] The abnormal pattern library is a pre-built database containing various known abnormal behavior patterns and characteristics. The library is used to match with real-time perception data to identify potential abnormal activities.

[0063] Dynamic matching processing refers to the process of comparing real-time environmental perception data with patterns in the abnormal pattern library, and automatically analyzing the similarity between the two through algorithms to determine whether there is potential abnormal activity.

[0064] The potential abnormal activity list is a list generated after dynamic matching processing, which lists all possible abnormal activities and their related information, including the time, location, type, etc.

[0065] In an embodiment of the present application, first, the system receives real-time environmental perception data from multiple sensors and monitoring devices; second, the system loads a predefined abnormal pattern library and prepares for matching; third, a dynamic matching algorithm is used to compare the real-time data with the abnormal pattern and calculate the similarity; finally, a list of potential abnormal activities is generated based on the matching results, and all suspected abnormal events are recorded.

[0066] Suppose that in an intelligent emergency evacuation cabin, the system needs to monitor the patient's real-time status and provide intelligent auxiliary treatment. First, a camera is installed in the cabin to monitor the patient's status during unmanned emergency transportation, and data analysis is performed based on the patient's facial expressions, body movements and other information; secondly, the system loads a pattern library containing a variety of abnormal physiological states (such as shortness of breath, confusion, limb convulsions, etc.); thirdly, a dynamic matching algorithm is used to compare real-time data with predefined patterns, calculate similarity scores, and identify potential emergency medical situations; finally, a list of potential abnormal activities is generated, listing all possible abnormal situations, such as sudden difficulty breathing or loss of consciousness of the patient.

[0067] 102. Based on the potential abnormal activity list, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and the key spatiotemporal feature points in the dynamic capture results are weighted by a feature weighting method to generate an abnormal behavior trend prediction model;

[0068] The spatiotemporal dynamic graph neural flow algorithm is a deep learning algorithm that combines the spatiotemporal dimension and graph structure. It can process complex data with temporal and spatial correlation. It captures the dynamic changes between nodes through the continuous time graph convolution mechanism.

[0069] The continuous-time graph convolution mechanism refers to the introduction of the time dimension into the graph neural network, which enables the model to capture the changes in the relationship between nodes at different time points, which helps to better understand the dynamic patterns in time series data.

[0070] Key spatiotemporal feature points refer to those data points that have significant influence in the time and space dimensions. They are usually important signs of abnormal behavior and are crucial for subsequent analysis.

[0071] The abnormal behavior trend prediction model is a mathematical model generated based on the spatiotemporal dynamic graph neural flow algorithm, which is used to predict the development trend of abnormal behavior. The model can help identify potential risks in advance and provide support for decision-making.

[0072] In the embodiment of the present application, firstly, the system applies the spatiotemporal dynamic graph neural flow algorithm to process the data in the list of potential abnormal activities; secondly, the time series data and spatial correlation of abnormal activities are captured through the continuous time graph convolution mechanism; thirdly, the feature weighting method is used to evaluate the importance and weight the key spatiotemporal feature points; finally, an abnormal behavior trend prediction model is generated to predict the future evolution of abnormal behavior.

[0073] For example, continuing with the above example, suppose that in an intelligent emergency evacuation cabin, the system has generated a list of potential abnormal activities, which includes abnormal physiological states of some patients (such as shortness of breath, confusion, etc.). Based on this list, the system first applies the spatiotemporal dynamic graph neural flow algorithm to process the relevant data; secondly, the continuous time graph convolution mechanism is used to capture the physiological change patterns and spatial distribution (such as facial expressions and limb movements) of these patients during the transportation process; thirdly, key spatiotemporal feature points are identified, such as the violent respiratory fluctuations or sudden limb twitches of the patient at a certain moment, and weighted processing is performed on them; finally, an abnormal behavior trend prediction model is generated to predict the possible deterioration of the patient's health in the future, and to assist in guiding the improvement of the oxygen concentration of the ventilator to enhance comfort, and to adjust the tidal volume and minute ventilation and other respiratory states.

[0074] 103. Based on the abnormal behavior trend prediction model, use the federated transfer learning algorithm based on differential privacy to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract the key features of the risk score, dynamically set the grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate abnormal behavior urgency;

[0075] In this step, differential privacy technology is a technology that protects personal privacy and ensures that the specific information of individuals will not be leaked during data analysis. By adding noise or other means, the impact of a single data point is blurred.

[0076] The federated transfer learning algorithm is a distributed machine learning method that allows different parties to jointly train models without sharing original data. Transfer learning can use the knowledge of existing models to accelerate the learning process of new tasks.

[0077] Risk scoring refers to a quantitative assessment of the output of the abnormal behavior trend prediction model to give a numerical value indicating the degree of risk.

[0078] Emergency grading analysis is to classify abnormal behaviors into different emergency levels based on risk scores so that appropriate response measures can be taken.

[0079] Causal inference is a statistical method used to analyze the cause-effect relationship between variables. This method can better understand the reasons behind abnormal behavior and develop more effective intervention strategies.

[0080] The urgency of abnormal behavior refers to the final risk assessment of abnormal behavior based on the results of the urgency classification analysis to determine its urgency.

[0081] In the embodiment of the present application, first, the system performs interval distribution analysis based on the risk score output by the abnormal behavior trend prediction model; secondly, the system uses a federated transfer learning algorithm based on differential privacy to perform urgency grading analysis on the risk score; thirdly, the key features of the risk score are extracted and the grading threshold is dynamically set; finally, causal inference technology is used to perform causal relationship analysis on the grading results, and the grading threshold is dynamically adjusted to generate the urgency of the abnormal behavior.

[0082] For example, continuing with the above example, assuming that the system has generated an abnormal behavior trend prediction model and predicted the possible health deterioration of a patient in the future. The system first performs interval distribution analysis based on the risk score output by the model to evaluate the risk probability of different score intervals; secondly, it uses the federated transfer learning algorithm based on differential privacy to perform urgency classification analysis on the risk score to ensure that the patient's privacy is not violated; thirdly, it extracts key features and dynamically sets the classification threshold to divide the risk score into different emergency levels; finally, it uses causal inference technology to perform causal relationship analysis on the classification results, dynamically adjusts the classification threshold, generates the final abnormal behavior urgency, and guides the medical team to take appropriate treatment measures, such as adjusting the injection pump for sedative injection for manic patients and adrenaline injection when necessary.

[0083] 104. Output a customized safety notification based on the urgency of the abnormal behavior and send it to the target source group of the real-time environmental perception data, continuously monitor the user behavior and environmental changes of the target source group, dynamically update the abnormal pattern library, and generate intelligent recognition and early warning strategies.

[0084] In this step, customized safety notifications are generated by the system for specific groups or scenarios based on the urgency of abnormal behavior.

[0085] The target source group refers to those people or units related to abnormal behavior, such as affected residents, enterprises, government departments, etc. They are the main recipients of customized security notifications.

[0086] Continuous monitoring means that after issuing a safety notification, the system continues to monitor the behavioral changes and environmental conditions of the target source group in order to evaluate the effectiveness of the early warning and adjust the strategy in a timely manner.

[0087] The intelligent identification and early warning strategy is to dynamically update the abnormal pattern library based on the results of continuous monitoring and optimize the system's identification capabilities and early warning mechanisms. This strategy ensures that the system can adapt to new threats and changes and maintain long-term effectiveness.

[0088] In the embodiment of the present application, first, the system generates customized security notifications based on the urgency of abnormal behavior; second, the notifications are sent to the target source group of real-time environmental perception data; third, the user behavior and environmental changes of the target source group are continuously monitored; finally, the abnormal pattern library is dynamically updated to generate intelligent identification and early warning strategies to ensure the adaptability and reliability of the system.

[0089] Continuing with the above example, assume that the system has generated an abnormal behavior urgency, indicating that a patient is about to experience a serious health deterioration. The system first generates a customized safety notification based on the urgency, informing the accompanying medical staff and the telemedicine center to pay attention to strengthening monitoring; secondly, the notification is sent to the relevant medical management department and the rescue team involved in the transfer; thirdly, the behavioral changes and environmental conditions of these groups are continuously monitored to evaluate the warning effect, such as real-time monitoring of the patient's facial expressions and body movements through the camera; finally, the abnormal pattern library is dynamically updated according to the monitoring results, and intelligent recognition and warning strategies are generated to ensure that the system can better respond to similar situations in the future, thereby improving the first aid effect and patient safety. In summary, steps 101 to 104 cover the complete process from receiving real-time environmental perception data and abnormal pattern matching, to dynamic capture and trend prediction of abnormal behavior, to urgency classification analysis of risk scoring, and finally outputting customized safety notifications and continuously optimizing warning strategies, aiming to provide a comprehensive and intelligent abnormal behavior recognition and warning system to meet the needs of public safety and management efficiency in smart cities and the Internet of Things environment.

[0090] In order to further improve the accuracy and comprehensiveness of abnormal behavior identification, in some embodiments, the use of the spatiotemporal dynamic graph neural flow algorithm based on the potential abnormal activity list in step 102 includes: based on the potential abnormal activity list, using multi-source information fusion technology to collect heterogeneous data in the potential abnormal activity list for integration and synchronization processing to generate a heterogeneous data set; based on the heterogeneous data set, using the spatiotemporal dynamic graph neural flow algorithm, through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list to generate an abnormal behavior feature map; based on the abnormal behavior feature map, using a feature weighting method to weight key spatiotemporal feature points in the dynamic capture results, identify feature points with important influence in the weighted processing results, and generate a key spatiotemporal feature set; based on the key spatiotemporal feature set, combined with the modeling capability of the spatiotemporal dynamic graph neural flow algorithm, model the time series characteristics of abnormal behavior in the abnormal behavior feature map to generate an abnormal behavior trend prediction model.

[0091] In this embodiment, multi-source information fusion technology refers to the technology of integrating and synchronously processing data from different sources (such as sensor data, video surveillance data, etc.). These heterogeneous data are cleaned and converted to form a unified data set for subsequent analysis.

[0092] A heterogeneous dataset is a data set consisting of multiple different types of data, such as time series data, image data, text data, etc. These data may come from different acquisition devices or systems and have different formats and structures.

[0093] The abnormal behavior feature map is a visualization tool generated by the spatiotemporal dynamic graph neural flow algorithm, which shows the spatiotemporal relationship between each feature point in abnormal activities. It helps to intuitively understand the development process of abnormal behavior and its spatial distribution.

[0094] Key spatiotemporal feature points are those data points in the abnormal behavior feature map that are crucial to the identification and prediction of abnormal behavior. They are usually located at key positions of abnormal behavior and show significant changes in time and space dimensions.

[0095] The key spatiotemporal feature set is a set of the most representative and influential feature points selected from all key spatiotemporal feature points. By weighting these feature points, the recognition ability and prediction accuracy of the model can be enhanced.

[0096] In the embodiment of the present application, firstly, the system collects and integrates heterogeneous data based on the data in the list of potential abnormal activities using multi-source information fusion technology to generate a comprehensive heterogeneous data set; secondly, the time series data and spatial correlation in the heterogeneous data set are dynamically captured using the spatiotemporal dynamic graph neural flow algorithm to generate a detailed abnormal behavior feature map; thirdly, for the key spatiotemporal feature points in the feature map, a feature weighting method is used to evaluate the importance, identify feature points with significant impact, and generate a key spatiotemporal feature set; finally, combined with the powerful modeling capability of the spatiotemporal dynamic graph neural flow algorithm, the time series characteristics of abnormal behavior are modeled to generate an accurate abnormal behavior trend prediction model.

[0097] Here is a specific example:

[0098] Assume that in an intelligent emergency evacuation cabin, the system needs to improve its ability to identify and warn patients of abnormal physiological states. First, a camera is installed in the cabin to monitor the patient's status during unmanned emergency transportation, and data analysis is performed based on the patient's facial expressions, limb behavior and other information; second, the system uses multi-source information fusion technology to collect data from a variety of medical devices such as ventilators, ECG monitors, and injection pumps to generate a heterogeneous data set containing information such as respiratory rate, heart rate, oxygen saturation, and drug dosage; third, the spatiotemporal dynamic graph neural flow algorithm is used to process these data to generate a detailed abnormal behavior feature map, showing the patient's physiological change pattern and its spatial distribution during the transportation process; finally, key spatiotemporal feature points are identified, such as the patient's violent respiratory fluctuations or sudden limb convulsions at a certain moment, and weighted processing is performed to generate a key spatiotemporal feature set.

[0099] In order to further improve the accuracy and data quality of abnormal behavior identification, in some embodiments, the use of the spatiotemporal dynamic graph neural flow algorithm based on heterogeneous data sets in step 102 includes: based on the heterogeneous data sets, using data cleaning and denoising technology to extract the time series data and spatial correlation information in the heterogeneous data sets for preprocessing to generate a preprocessed data set; based on the preprocessed data set, using the spatiotemporal dynamic graph neural flow algorithm, through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list, and generate a dynamic capture result; based on the dynamic capture result, applying pattern recognition technology, classifying and clustering the abnormal patterns in the dynamic capture results, and generating an abnormal pattern library; based on the abnormal pattern library, visually analyzing the abnormal behaviors in the abnormal pattern library by constructing a multi-dimensional icon, displaying the spatiotemporal distribution characteristics of the abnormal behaviors, and generating an abnormal behavior feature map.

[0100] In this embodiment, data cleaning and denoising technology refers to removing noise and incomplete information in the data through a series of processing methods to ensure the accuracy and consistency of the data. Commonly used methods include filtering, smoothing, missing value filling, etc.

[0101] The preprocessed data set is a data set that has been cleaned and denoised, retaining the core features of the original data while reducing interference factors.

[0102] Time series data refers to data points arranged in chronological order, usually used to describe the change of a variable over time.

[0103] Spatial correlation information refers to the relationship between different locations, such as the impact of traffic flow between two adjacent intersections.

[0104] The dynamic capture results are generated by capturing the time series data and spatial correlation of abnormal activities through the spatiotemporal dynamic graph neural flow algorithm, reflecting the development process of abnormal behavior and its spatiotemporal characteristics.

[0105] Pattern recognition technology is a technology that automatically discovers and classifies patterns from data and is widely used in image recognition, speech recognition and other fields.

[0106] The abnormal pattern library is a database that stores various known abnormal patterns, each of which contains its characteristic description and classification label.

[0107] Multidimensional icons are a visualization tool that can graphically display multiple dimensions of complex data, helping users intuitively understand the correlation and changing trends between data.

[0108] The abnormal behavior feature map is a chart generated by visually analyzing the abnormal behaviors in the abnormal pattern library to show the spatiotemporal distribution characteristics of the abnormal behaviors and provide support for decision-making.

[0109] In the embodiment of the present application, firstly, the system is based on heterogeneous data sets, and uses data cleaning and denoising technology to extract the time series data and spatial correlation information therein for preprocessing, so as to generate a high-quality preprocessed data set; secondly, the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the abnormal activity time series data and spatial correlation in the preprocessed data set, and generate detailed dynamic capture results; thirdly, the pattern recognition technology is used to classify and cluster the abnormal patterns in the dynamic capture results, and a comprehensive abnormal pattern library is constructed; finally, a multi-dimensional icon is constructed to visualize the abnormal behaviors in the abnormal pattern library, display the spatiotemporal distribution characteristics of these behaviors, and generate an abnormal behavior feature map.

[0110] Here is a specific example:

[0111] Assume that in an intelligent emergency transfer cabin, the system needs to improve its ability to identify potential health risks. First, cameras and various medical sensors are installed in the cabin to monitor the patient status during unmanned emergency transfer, and data analysis is performed through the patient's facial expressions, body movements and other information; secondly, based on the collected heterogeneous data sets (such as respiratory rate, heart rate, oxygen saturation, drug dosage, etc.), the system uses data cleaning and denoising technology to extract the time series data and spatial correlation information from them for preprocessing and generate a preprocessed data set; thirdly, the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal physiological activities in the preprocessed data set, and generate detailed dynamic capture results; then, pattern recognition technology is used to classify and cluster the abnormal patterns in these dynamic capture results, and an abnormal pattern library containing multiple abnormal physiological states is constructed; finally, a multi-dimensional chart is constructed to visualize the abnormal behaviors in the abnormal pattern library, display the spatiotemporal distribution characteristics of these behaviors, and generate abnormal behavior feature maps to provide intuitive early warning information and decision support for medical staff.

[0112] In order to further improve the accuracy of abnormal behavior identification and the importance of feature points, in some embodiments, the feature weighting method based on the abnormal behavior feature map in step 102 includes: based on the abnormal behavior feature map, extracting and processing representative spatiotemporal 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, using a feature weighting method to weight key spatiotemporal feature points in the dynamic capture result, assigning different weights to different feature points in the key spatiotemporal feature points to distinguish the importance of the different feature points, and generating a weighted feature point set; based on the weighted feature point set, using association rule mining technology to extract the intrinsic connection of the spatiotemporal features of different weighted feature points in the weighted feature point set to generate a spatiotemporal association rule base; based on the spatiotemporal association rule base, extracting relevant spatiotemporal feature points in the spatiotemporal association rule base, performing feature point combination processing, and generating a key spatiotemporal feature set.

[0113] In this embodiment, the feature point extraction method refers to a technique for identifying representative spatiotemporal feature points from a data set. These feature points are usually data points that show significant changes or patterns in a specific time and space. Common extraction methods include threshold segmentation, edge detection, and the like.

[0114] The preliminary feature point set is a set of initial feature points generated by the feature point extraction method. These points may contain redundant information and need to be further processed to improve their representativeness.

[0115] The feature weighting method is a method of assigning different weights to different feature points to distinguish the importance of each feature point. By calculating the contribution of each feature point, their role in abnormal behavior can be more accurately reflected.

[0116] The weighted feature point set is a feature point set generated after feature weighting processing, in which each feature point is attached with a corresponding weight value.

[0117] Association rule mining technology is a method to discover the intrinsic connections between variables from large amounts of data. It can identify the temporal and spatial correlations between feature points and reveal hidden patterns.

[0118] The spatiotemporal association rule base is a database that stores various spatiotemporal association rules, and each rule describes the relationship between different feature points.

[0119] The key spatiotemporal feature set is a set of key feature points generated by combining and processing related spatiotemporal feature points in the spatiotemporal association rule base.

[0120] In the embodiment of the present application, firstly, the system extracts and processes representative spatiotemporal feature points based on the abnormal behavior feature map through a feature point extraction method to generate a preliminary feature point set; secondly, a feature weighting method is used to weight the key spatiotemporal feature points in the dynamic capture results, and different weights are assigned according to the importance of different feature points to generate a weighted feature point set; thirdly, the spatiotemporal features of different weighted feature points in the weighted feature point set are extracted and processed through association rule mining technology to generate a spatiotemporal association rule base; finally, based on the spatiotemporal association rule base, the relevant spatiotemporal feature points are extracted, and the feature point combination processing is performed to generate a key spatiotemporal feature set.

[0121] Here is a specific example:

[0122] Suppose that in an intelligent emergency transport cabin, the system needs to accurately identify potential health risks. First, cameras and various medical sensors are installed in the cabin to monitor the patient status during unmanned emergency transport, and data analysis is performed through information such as the patient's facial expressions, body movements, etc.; secondly, based on the abnormal physiological feature map, the system extracts and processes representative spatiotemporal feature points through feature point extraction methods to generate a preliminary feature point set, such as heart rate fluctuations or respiratory rate changes at a certain moment; thirdly, the feature weighting method is used to weight these key spatiotemporal feature points, and different weights are assigned according to the importance of different feature points (such as heart rate fluctuation amplitude, degree of shortness of breath), to generate a weighted feature point set; then, the spatiotemporal features of different weighted feature points in the weighted feature point set are extracted and processed through association rule mining technology to generate a spatiotemporal association rule base, revealing the potential connections between different feature points; finally, based on the spatiotemporal association rule base, relevant spatiotemporal feature points are extracted, such as the frequent occurrence of severe respiratory fluctuation areas in a certain period of time, and feature point combination processing is performed to generate a key spatiotemporal feature set to help medical staff quickly locate and respond to abnormal situations.

[0123] In order to further improve the accuracy of abnormal behavior identification and the flexibility of response, in some embodiments, the federated transfer learning algorithm based on differential privacy is used based on the abnormal behavior trend prediction model in step 103, including: based on the abnormal behavior trend prediction model, performing interval distribution analysis on the risk score output by the abnormal behavior trend prediction model, evaluating the risk probability of different score intervals in the interval distribution analysis, and generating a risk probability distribution map; based on the risk probability distribution map, using the federated transfer learning algorithm based on differential privacy, performing urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extracting the key features of the risk score, dynamically setting the grading threshold of the risk score, and generating a preliminary urgency grading result; based on the preliminary urgency grading result, using causal inference technology, performing causal relationship analysis on the urgency grading analysis result, dynamically adjusting the grading threshold, and generating an optimized urgency grading result; based on the optimized urgency grading result, establishing a multi-level feedback mechanism, setting different feedback paths for different levels of urgency in the optimized urgency grading result, and generating abnormal behavior urgency.

[0124] In this embodiment, interval distribution analysis refers to segmenting the risk score and evaluating the probability distribution of each score interval. This analysis can better understand the risk levels of different score intervals and provide a basis for subsequent classification.

[0125] The risk probability distribution diagram is a visualization tool that shows the probability distribution of different risk score intervals. It helps to intuitively identify high-risk and low-risk areas and guide urgency grading analysis.

[0126] Emergency grading analysis is to classify abnormal behaviors into different emergency levels according to risk scores so that appropriate response measures can be taken.

[0127] Dynamically set classification thresholds to continuously adjust risk score classification standards based on actual conditions to adapt to changing environments and new threats.

[0128] The optimization of the urgency grading results is to conduct a causal analysis on the preliminary urgency grading results, dynamically adjust the grading thresholds, and generate a more scientific and reasonable final urgency grading result.

[0129] The multi-level feedback mechanism is a multi-level feedback system that sets different feedback paths for different levels of urgency.

[0130] In the embodiment of the present application, first, the system performs an interval distribution analysis on the output risk score based on the abnormal behavior trend prediction model, evaluates the risk probability of different score intervals, and generates a detailed risk probability distribution diagram; secondly, the system uses a federated transfer learning algorithm based on differential privacy to perform an urgency grading analysis on the risk score, extracts key features and dynamically sets the grading threshold, and generates a preliminary urgency grading result; thirdly, the causal inference technology is used to perform a causal relationship analysis on the preliminary urgency grading result, dynamically adjusts the grading threshold, and generates an optimized urgency grading result; finally, based on the optimized urgency grading result, a multi-level feedback mechanism is established, different feedback paths are set for different levels of urgency, and the final abnormal behavior urgency is generated.

[0131] Here is a specific example:

[0132] Suppose in an intelligent emergency transport system, the system needs to accurately assess and respond to potential health risks in a timely manner. First, cameras and various medical sensors are installed in the cabin to monitor the patient status during unmanned emergency transport, and perform data analysis based on the patient's facial expressions, limb behaviors and other information; secondly, based on the abnormal physiological trend prediction model, the system performs interval distribution analysis on the output risk score, evaluates the risk probability of different score intervals, and generates a detailed risk probability distribution map, for example, showing which patients may experience breathing difficulties or abnormal heart rate in the future; thirdly, using the federated transfer learning algorithm based on differential privacy, the risk score is analyzed for urgency grading, key features such as heart rate fluctuations, respiratory rate changes, etc. are extracted, and dynamically set The system uses causal inference technology to conduct causal relationship analysis on the preliminary urgency grading results, for example, to explore whether drastic respiratory fluctuations within a specific time period lead to a higher risk of worsening of the disease, dynamically adjust the grading threshold, and generate optimized urgency grading results. Finally, based on the optimized urgency grading results, a multi-level feedback mechanism is established to set different feedback paths for different levels of urgency, such as sending alarms to accompanying medical staff or automatically adjusting the parameters of medical equipment in the cabin (such as increasing the oxygen concentration of the ventilator to enhance comfort, adjusting the tidal volume and minute ventilation), to generate the final urgency of abnormal behavior and ensure timely and effective response measures.

[0133] In order to further improve the accuracy of abnormal behavior identification and the flexibility of response, in some embodiments, the method of using a federated transfer learning algorithm based on differential privacy based on the risk probability distribution map in step 103 includes: conducting in-depth analysis based on the risk probability distribution map, focusing on the overall risk pattern and trend in the risk probability distribution map, vectorizing the overall risk pattern and trend, and generating a risk feature matrix; based on the risk feature matrix, applying a federated transfer learning algorithm based on differential privacy to each risk feature vector in the risk feature matrix, performing an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, and generating an urgency grading score; based on the urgency grading score, performing score fusion processing by extracting key features of the risk score, dynamically setting the grading threshold of the risk score according to the score fusion processing result, and generating an optimized grading threshold; based on the optimized grading threshold, introducing an adaptive adjustment mechanism, recording the output data in real time during the operation of the abnormal behavior trend prediction model, and performing a comparative analysis with the optimized grading threshold for automatic adjustment to generate a preliminary urgency grading result.

[0134] In this embodiment, the risk probability distribution diagram is a visualization tool that shows the probability distribution of different risk score intervals.

[0135] Overall risk patterns and trends refer to the global risk characteristics reflected in the risk probability distribution diagram, including the changing trends, peaks and troughs of risk levels.

[0136] The risk feature matrix is ​​a matrix composed of multiple risk feature vectors, each of which represents a risk feature at a specific time point or spatial location.

[0137] The urgency rating is the result of grading the abnormal behavior risk score, reflecting the urgency of different events. The higher the score, the more urgent the event is and needs to be handled with priority.

[0138] Scoring fusion processing combines the results of multiple scoring methods to generate more comprehensive and accurate scoring results.

[0139] Optimizing the classification threshold means continuously adjusting the risk score classification standards according to actual conditions to adapt to changing environments and new threats.

[0140] The adaptive adjustment mechanism is a mechanism that can automatically adjust system parameters according to real-time data. It can continuously optimize the performance of the system during operation to ensure that it is always in the best state.

[0141] In the embodiment of the present application, first, the system conducts in-depth analysis based on the risk probability distribution diagram, focuses on the overall risk patterns and trends, performs vector processing on them, and generates a risk feature matrix; secondly, the system utilizes a federated transfer learning algorithm based on differential privacy to perform urgency grading analysis on each risk feature vector in the risk feature matrix, and generates an urgency grading score; thirdly, the key features of the urgency grading score are scored and fused, and the grading threshold of the risk score is dynamically set according to the fusion processing result to generate an optimized grading threshold; finally, an adaptive adjustment mechanism is introduced to record the output data in real time during the operation of the abnormal behavior trend prediction model, and compare and analyze it with the optimized grading threshold for automatic adjustment to generate a preliminary urgency grading result.

[0142] Here is a specific example:

[0143] Assume that in an intelligent emergency rescue cabin, the system needs to accurately assess and respond to potential health risks in a timely manner. First, based on the data collected from cameras and medical sensors, the system generates a risk feature matrix, focusing on overall health patterns and trends, such as the changing trend of patients' vital signs over a period of time, and quantizes them to provide a basis for subsequent analysis. Second, using the federated transfer learning algorithm based on differential privacy, the urgency grading analysis of each risk feature vector in the risk feature matrix is ​​performed, such as evaluating the health risk level in different time periods, generating an urgency grading score, and ensuring that the patient's condition is accurately assessed. Third, by fusion processing of the key features of the urgency grading score (such as heart rate, respiratory rate, and blood oxygen saturation), the grading threshold of the risk score is dynamically set according to the fusion processing results, and the optimized grading threshold is generated to more accurately reflect the patient's current health status. Finally, an adaptive adjustment mechanism is introduced to record the output data in real time during the operation of the abnormal physiological trend prediction model, such as the newly emerging vital sign abnormalities, and compare and analyze them with the optimized grading threshold to automatically adjust and generate a preliminary urgency grading result. This ensures that the system can respond quickly and effectively to sudden changes in a patient's health. For example, when a patient is detected to be experiencing manic symptoms, the corresponding sedation procedure can be immediately initiated.

[0144] In order to further improve the response efficiency of abnormal behavior identification and the intelligence of early warning strategies, in some embodiments, the output of customized security notifications based on the urgency of abnormal behavior in step 104 includes: outputting customized security notifications based on the urgency of abnormal behavior, and sending them to the target source group of the real-time environmental perception data to generate a customized security notification record; setting an automated response process based on the customized security notification record to automate the response process of the target source group after receiving the customized security notification record to generate an automated response log; based on the automated response log, continuously monitoring user behaviors and environmental changes of the target source group to capture potential abnormal activities during the continuous monitoring process and generate a real-time monitoring report; based on the real-time monitoring report, extracting newly generated abnormal patterns in the real-time monitoring report to dynamically update the abnormal pattern library to generate intelligent identification and early warning strategies.

[0145] In this embodiment, the customized safety notification is that the system generates safety warning information for specific groups or scenarios according to the different urgency of abnormal behavior.

[0146] Customized security notification records refer to the notification details automatically recorded by the system after generating and sending customized security notifications, including the recipient, notification time, content summary, etc.

[0147] An automated response process is a series of pre-set automated actions that ensure that the target source group can take appropriate measures quickly after receiving customized security notifications.

[0148] The automated response log records the execution status of the automated response process, including response time, execution results, and other information.

[0149] Real-time monitoring reports are reports generated by the continuous monitoring process, which record in detail the user behavior and environmental changes of the target source group, as well as potential abnormal activities captured in the process.

[0150] Newly generated abnormal patterns refer to new abnormal behavior patterns found in real-time monitoring reports that have never appeared in the abnormal pattern library.

[0151] The intelligent identification and early warning strategy is to dynamically update the abnormal pattern library based on real-time monitoring reports and newly generated abnormal patterns, and optimize the system's identification capabilities and early warning mechanisms. This strategy ensures that the system can adapt to new threats and changes and maintain long-term effectiveness.

[0152] In an embodiment of the present application, first, the system outputs a customized security notification based on the urgency of the abnormal behavior, and sends it to the target source group of the real-time environmental perception data, and generates a customized security notification record at the same time; secondly, based on the customized security notification record, an automated response process is set to ensure that the response process of the target source group after receiving the notification is automated, and an automated response log is generated; thirdly, based on the automated response log, the user behavior and environmental changes of the target source group are continuously monitored to capture potential abnormal activities during the continuous monitoring process and generate a real-time monitoring report; finally, based on the real-time monitoring report, the newly generated abnormal patterns are extracted to dynamically update the abnormal pattern library and generate intelligent identification and early warning strategies.

[0153] Here is a specific example:

[0154] Suppose in an intelligent multifunctional emergency transfer cabin, the system needs to improve the response speed and warning accuracy to patient health risks. First, cameras and various medical sensors are installed in the cabin to monitor the patient's status during unmanned emergency treatment. Data analysis is performed by analyzing the patient's facial expressions, body movements and other information to assist in guiding the adjustment of ventilator parameters to enhance comfort and optimize respiratory support (such as increasing oxygen concentration, adjusting tidal volume and minute ventilation), and emergency treatment such as sedative push or epinephrine push is performed by adjusting the syringe pump when necessary. Secondly, customized safety notifications are output based on the abnormal physiological urgency and sent to accompanying medical staff and telemedicine centers. At the same time, customized safety notification records are generated, such as an early warning that a patient's condition is about to deteriorate seriously. Thirdly, based on the customized safety notification records, an automated response process is set up, such as automatically adjusting the parameters of medical equipment in the cabin, starting emergency treatment procedures, and generating automated response logs. Finally, based on the automated response logs, the patient's behavioral changes and vital signs are continuously monitored to capture potential signs of health deterioration and generate real-time monitoring reports to ensure that the system can adjust the treatment plan in a timely manner and dynamically update the abnormal pattern library, generate intelligent recognition and early warning strategies, and ensure that the system can better cope with similar situations in the future.

[0155] This application considers that in order to solve the problems in the prior art of insufficient abnormal behavior recognition accuracy, untimely response, and limited processing capabilities for complex spatiotemporal correlations, the invention embodiment proposes this optional solution to solve these problems, and thus proposes a new optional solution, which includes:

[0156] Based on the preprocessed data set, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and generate dynamic capture results, including:

[0157] Based on the preprocessed data set, short-term fluctuations are eliminated by using time series smoothing technology, spatial coordinate normalization methods are applied to unify spatial coordinates of different scales to the same magnitude, and abnormal activity intensity is standardized to generate preliminary characteristic response values;

[0158] The preliminary characteristic response value is calculated by the following formula:

[0159] R(t)=η·exp(-λ·|T(t)-T0|-v·|S(t)-S0|)+k·|A(t)-A0|

[0160] Where R(t) is the initial characteristic response value of the time series data of abnormal activity at time t and the spatial correlation; η is the response intensity coefficient; λ is the time deviation attenuation factor; T(t) and T0 are the timestamps of time t and the reference time, respectively; v is the position deviation correction coefficient; S(t) and S0 are the spatial coordinates of time t and the reference time, respectively; κ is the activity intensity weight; A(t) and A0 are the abnormal activity intensities at time t and the reference time, respectively; t is a time variable, which is used to refer to a specific time;

[0161] Based on the preliminary characteristic response values, the characteristic value clustering analysis is used to identify the node groups with similar characteristic response patterns, the distinction between features is enhanced through nonlinear transformation, the time and space periodicity analysis is combined to capture the potential periodic abnormal behavior pattern, and a weighted comprehensive evaluation mechanism is introduced to generate a comprehensive dynamic capture index;

[0162] The comprehensive dynamic capture index is calculated by the following formula:

[0163]

[0164] Where C(t) is the comprehensive dynamic capture index at time t; i is the node index involved in the calculation, from 1 to n; n is the number of nodes involved in the calculation; α i is the response adjustment coefficient of the i-th node; β i is the nonlinear response adjustment factor of the ith node; R(t) is the initial characteristic response value at time t; γ i is the time period coefficient of the i-th node; δ i is the time frequency factor of the i-th node; T(t) and T0 are the time of time t and the reference time respectively; θ i is the position period coefficient of the i-th node; ∈ i is the position frequency factor of the i-th node; S(t) and S0 are the spatial coordinates at time t and the reference time respectively;

[0165] Based on the comprehensive dynamic capture index, through threshold screening and outlier detection, abnormal events that significantly deviate from the normal range are identified and classified and labeled, and the time series prediction model is used to estimate the future abnormal behavior trend, provide early warning information, and generate dynamic capture results.

[0166] This method aims to improve the accuracy and response efficiency of abnormal behavior identification. It quantifies the impact of time series data and spatial correlation of abnormal activities through preliminary feature response values, and further enhances the distinction between features through nonlinear transformation and periodic analysis based on the preliminary feature response values ​​through a comprehensive dynamic capture index. It improves the system's recognition ability and warning accuracy, which not only solves the problems in the existing technology, but also provides a more scientific and effective solution for the field of intelligent identification and warning.

[0167] Assume that in an intelligent emergency medical cabin, the system needs to monitor the patient's health status in real time and warn of potential health deterioration in advance. ; Assume that the timestamp T(t) = 1678456789, the reference time T0 = 1678456780; the spatial coordinates S(t) = (40.7128, -74.0060), the reference coordinates S0 = (40.7128, -74.0059); the abnormal activity intensity A(t) = 0.85, the reference intensity A0 = 0.5; the coefficient η = 1.0, λ = 0.01, v = 0.02, κ = 0.5;

[0168] R(t)=

[0169] 1.0·exp(-0.01·|1678456789-1678456780|-0.02·|40.7128-40.7128|)+0.5·∣

[0170] 0.85-0.5|=0.775;

[0171] Assume that the number of nodes involved in the calculation is n = 3; for the first node, set α1 = 0.8, β1 = 0.05, γ1 = 0.6, δ1 = 0.01, θ1 = 0.4, ∈1 = 0.02; set other node parameters similarly;

[0172]

[0173] Assume that the threshold is set to 2. Since the calculated result 2.15 is greater than the set threshold, it indicates that the patient's health may seriously deteriorate in the future. This is because the higher comprehensive dynamic capture index reflects the system's high sensitivity to changes in the patient's condition, and also verifies the effectiveness of abnormal behavior recognition and early warning. Through the above steps, real-time monitoring and early warning of the patient's status are ensured, the efficiency and scientificity of emergency medical care are improved, the response speed and reliability of the entire system are enhanced, and the treatment plan is adjusted in time, thereby improving the first aid effect and patient safety. The present application takes into account that in order to solve the problems in the prior art due to insufficient accuracy in abnormal behavior recognition, inaccurate risk assessment, and limited ability to process complex spatiotemporal correlations, the invention embodiment proposes this optional solution to solve these problems. Therefore, a new optional solution is proposed, which includes:

[0174] Based on the risk feature matrix, a federated transfer learning algorithm based on differential privacy is applied to each risk feature vector in the risk feature matrix, and an urgency grading analysis is performed on the risk score output by the abnormal behavior trend prediction model to generate an urgency grading score, including:

[0175] Based on the risk feature matrix, the risk features of different scales in the risk feature matrix are standardized through standardization processing, the main feature vectors in the standardized result are extracted by principal component analysis to reduce the data dimension and retain the most significant information, and the outlier detection technology is used to identify the outliers in the main feature vectors to generate a risk feature weighted index;

[0176] The risk characteristic weighted index is calculated using the following formula:

[0177]

[0178] Among them, W j is the risk characteristic weighted index of the j-th risk characteristic vector; ζ is the comprehensive adjustment coefficient; X p is the time series weight factor of the pth feature; R jp and R 0p are the time series data of the jth and the benchmark risk feature vectors at the pth feature respectively; ψ is the feature deviation correction coefficient; φ is the feature impact factor; F jp and F 0p are the spatial correlation features of the jth and the baseline risk feature vectors at the pth feature, respectively; p is the index of the feature, from 1 to m; m is the number of features;

[0179] Based on the risk feature weighted index, cluster analysis is used to group nodes with similar feature weighted indexes in the risk feature weighted index into one group, differences between different node groups are amplified through nonlinear transformation, and a time series prediction model is introduced to capture the time-varying trend of differences between the node groups to generate a risk response value;

[0180] The risk response value is calculated using the following formula:

[0181]

[0182] Among them, H i is the risk response value of the i-th risk feature vector; r is the index of the feature dimension involved in the calculation, from 1 to l; l is the number of feature dimensions involved in the calculation; α r is the response adjustment coefficient of the rth feature dimension; r is the nonlinear adjustment factor of the rth feature dimension; W i is the risk characteristic weighted index of the i-th risk characteristic vector; β r is the time series difference weight of the rth feature dimension; δ r is the time frequency factor of the rth feature dimension; C ir and C 0r are the time series characteristics of the i-th and benchmark risk feature vectors in the r-th feature dimension respectively; θ ris the position period coefficient of the rth feature dimension; ∈ r is the position frequency factor of the rth feature dimension; T ir and T 0r are the timestamps of the i-th and baseline risk feature vectors in the r-th feature dimension respectively; ρ r is the spatial influence coefficient of the rth feature dimension; γ r is the spatial sensitivity factor of the rth feature dimension; S ir and S 0r are the spatial coordinates of the ith and baseline risk feature vectors in the rth feature dimension, respectively;

[0183] Further threshold screening and classification annotation are performed based on the risk response value, high-risk nodes in the threshold screening and classification annotation results are identified for urgency grading analysis, and an adaptive threshold adjustment mechanism is introduced to generate an urgency grading score.

[0184] This method aims to improve the accuracy of abnormal behavior identification and the effectiveness of risk assessment by quantifying the importance of risk feature vectors, reducing data dimensions through standardization and principal component analysis, retaining the most significant information, and amplifying the differences between different node groups through cluster analysis and nonlinear transformation on the basis of risk feature weighted index. A time series prediction model is introduced to capture time-varying trends, thereby generating more accurate risk assessment results.

[0185] In the risk characteristic weighted index, the time series distance term This sub-item measures the distance of the risk feature vector in the time series, and enhances the sensitivity to small distance changes in the form of inverse square root; spatial correlation bias This sub-item reflects the deviation of spatial correlation features, amplifying small differences between features through logarithmic functions and product forms;

[0186] Among them, R jp and R 0p are the time series data of the j-th risk feature vector and the benchmark risk feature vector at the p-th feature, which can be obtained through historical data or real-time monitoring; F jp and F 0p are the spatial correlation features of the j-th risk feature vector and the benchmark risk feature vector in the p-th feature, which can be obtained through spatial coordinates or other location information; p ,ζ, ψ and φ are pre-set coefficients, usually determined based on statistical analysis of historical data or expert experience;

[0187] In the risk response value, the nonlinear response adjustment term This sub-item enhances the impact of the risk characteristic weighted index through the hyperbolic sine function and the exponential decay function, and takes into account the differences in time series characteristics; the time period term θ r ·arctan(∈ r ·(T ir -T 0r )):This item captures the periodic changes in time and reflects the difference between timestamps through the arc tangent function; spatial impact item This sub-item reflects the difference between spatial coordinates and amplifies the effect of spatial distance by the fourth power;

[0188] Among them, W i Calculated by the risk characteristic weighted index formula; C ir and C 0r are the time series characteristics of the i-th risk feature vector and the benchmark risk feature vector in the r-th feature dimension, which can be obtained through historical data or real-time monitoring; T ir and T 0r are the timestamps of the i-th risk feature vector and the benchmark risk feature vector in the r-th feature dimension, which can be obtained through the system built-in clock or external time source synchronization; S ir and S 0r are the spatial coordinates of the i-th risk feature vector and the benchmark risk feature vector in the r-th feature dimension, which can be obtained through GPS or other positioning devices; α r ,λ r ,β r ,δ r ,θ r ,∈ r ,ρ r and γ r It is a parameter set for each feature dimension, usually set through machine learning algorithm training or expert experience;

[0189] Assume that in an intelligent emergency medical cabin, the system needs to monitor potential health risks in real time and issue early warnings; Assume that the time series data R jp =[0.85,0.9], benchmark time series data R 0p =[0.6,0.7]; spatial correlation feature F jp =[0.75,0.7], benchmark spatial correlation feature F 0p =[0.55,0.6]; comprehensive adjustment coefficient ζ = 1.2; time series weight factor χ p =[0.9,0.8]; characteristic deviation correction coefficient ψ = 0.6; characteristic influence factor φ = 0.3;

[0190]

[0191] Assume that the number of feature dimensions involved in the calculation is l = 3; for the first feature dimension, set α1 = 0.7, λ1 = 0.04, β1 = 0.02, δ1 = 0.03, θ1 = 0.5, ∈1 = 0.04, ρ1 = 0.6, γ1 = 0.05;

[0192]

[0193] Assuming the threshold is set to 1.1, since the calculated result 1.32 is greater than the set threshold, it means that the patient has a high health risk, reflecting the system's high sensitivity to the patient's condition and verifying the effectiveness of the early warning mechanism. Through the above steps, real-time monitoring and early warning of potential health risks are ensured, the efficiency and scientificity of the emergency medical cabin in responding to emergencies are improved, the response speed and reliability of the entire system are enhanced, and the medical team can intervene in treatment earlier, improving the patient's chance of survival and the quality of recovery. This intelligent solution can not only respond quickly to sudden health changes in patients, but also help optimize the allocation of medical resources to ensure the best treatment effect. Figure 2 A schematic diagram of the structure of an intelligent identification and early warning system is provided for the embodiment of the present application. Figure 2 As shown, the device comprises:

[0194] A receiving module 21 is used to receive real-time environment perception data and predefine an abnormal pattern library, and based on the abnormal pattern library, dynamically match the real-time environment perception data to generate a potential abnormal activity list;

[0195] A processing module 22 is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list based on the potential abnormal activity list using a spatiotemporal dynamic graph neural flow algorithm through a continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and to weight key spatiotemporal feature points in the dynamic capture results using a feature weighting method to generate an abnormal behavior trend prediction model;

[0196] An analysis module 23 is used to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model based on the abnormal behavior trend prediction model by using a federated transfer learning algorithm based on differential privacy, extract key features of the risk score, dynamically set a grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate abnormal behavior urgency;

[0197] The monitoring module 24 is used to output customized safety notifications based on the urgency of the abnormal behavior, send them to the target source group of the real-time environmental perception data, continuously monitor the user behavior and environmental changes of the target source group, dynamically update the abnormal pattern library, and generate intelligent recognition and early warning strategies.

[0198] Figure 2 The intelligent identification and early warning system can be implemented Figure 1 The implementation principle and technical effect of the intelligent identification and early warning method described in the embodiment shown are not repeated here. The specific way in which each module and unit performs operations in the intelligent identification and early warning system in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0199] In one possible design, Figure 2 An intelligent identification and early warning system of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0200] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0201] The processing component 32 is used to: receive real-time environmental perception data and predefine an abnormal pattern library, based on the abnormal pattern library, dynamically match the real-time environmental perception data to generate a potential abnormal activity list; based on the potential abnormal activity list, use the spatiotemporal dynamic graph neural flow algorithm, through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list, use the feature weighting method to weight the key spatiotemporal feature points in the dynamic capture results, and generate an abnormal behavior trend prediction model; based on the abnormal behavior trend prediction model , use the federated transfer learning algorithm based on differential privacy to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract the key features of the risk score, dynamically set the grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate the urgency of abnormal behavior; output customized security notifications based on the urgency of abnormal behavior, send them to the target source group of the real-time environmental perception data, continuously monitor the user behavior and environmental changes of the target source group, dynamically update the abnormal pattern library, and generate intelligent recognition and early warning strategies.

[0202] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by 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 components to perform the above method.

[0203] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, 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 disk or optical disk.

[0204] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0205] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0206] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0207] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0208] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 An intelligent identification and early warning method of the illustrated embodiment.

[0209] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0210] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0211] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent identification and early warning method, characterized in that: include: Receive real-time environment perception data and predefine an abnormal pattern library, and based on the abnormal pattern library, dynamically match the real-time environment perception data to generate a potential abnormal activity list; Based on the potential abnormal activity list, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and the key spatiotemporal feature points in the dynamic capture results are weighted by a feature weighting method to generate an abnormal behavior trend prediction model; Based on the abnormal behavior trend prediction model, a federated transfer learning algorithm based on differential privacy is used to perform an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate the abnormal behavior urgency; Based on the urgency of the abnormal behavior, a customized safety notification is output and sent to the target source group of the real-time environmental perception data, the user behavior and environmental changes of the target source group are continuously monitored, the abnormal pattern library is dynamically updated, and an intelligent recognition and early warning strategy is generated.

2. The method according to claim 1, characterized in that Based on the potential abnormal activity list, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, and the key spatiotemporal feature points in the dynamic capture results are weighted by a feature weighting method to generate an abnormal behavior trend prediction model, including: Based on the potential abnormal activity list, using multi-source information fusion technology, collecting heterogeneous data in the potential abnormal activity list for integration and synchronization processing to generate a heterogeneous data set; Based on the heterogeneous data set, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm to generate an abnormal behavior feature map; Based on the abnormal behavior feature map, a feature weighting method is used to perform weighted processing on key spatiotemporal feature points in the dynamic capture results, and feature points with important influence in the weighted processing results are identified to generate a key spatiotemporal feature set; Based on the key spatiotemporal feature set and combined with the modeling capability of the spatiotemporal dynamic graph neural flow algorithm, the time series characteristics of the abnormal behavior in the abnormal behavior feature map are modeled to generate an abnormal behavior trend prediction model.

3. The method according to claim 2, characterized in that Based on the heterogeneous data set, the spatiotemporal dynamic graph neural flow algorithm is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm to generate abnormal behavior feature maps, including: Based on the heterogeneous data set, using data cleaning and denoising technology, extracting time series data and spatial correlation information in the heterogeneous data set for preprocessing to generate a preprocessed data set; Based on the preprocessed data set, using the spatiotemporal dynamic graph neural flow algorithm, through the continuous time graph convolution mechanism in the spatiotemporal dynamic graph neural flow algorithm, the time series data and spatial correlation of the abnormal activities in the abnormal activity list are dynamically captured to generate a dynamic capture result; Based on the dynamic capture results, applying pattern recognition technology, classifying and clustering the abnormal patterns in the dynamic capture results to generate an abnormal pattern library; Based on the abnormal pattern library, a multi-dimensional icon is constructed to perform a visual analysis of the abnormal behaviors in the abnormal pattern library, display the spatiotemporal distribution characteristics of the abnormal behaviors, and generate an abnormal behavior feature map.

4. The method according to claim 2, characterized in that: Based on the abnormal behavior feature map, a feature weighting method is used to perform weighted processing on key spatiotemporal feature points in the dynamic capture results, identify feature points with important influence in the weighted processing results, and generate a key spatiotemporal feature set, including: Based on the abnormal behavior feature map, extracting and processing representative spatiotemporal 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, a feature weighting method is used to perform weighted processing on the key spatiotemporal feature points in the dynamic capture result, and different weights are assigned to different feature points in the key spatiotemporal feature points to distinguish the importance of the different feature points, thereby generating a weighted feature point set; Based on the weighted feature point set, the temporal and spatial features of different weighted feature points in the weighted feature point set are extracted and processed by association rule mining technology to generate a temporal and spatial association rule library; Based on the spatiotemporal association rule base, relevant spatiotemporal feature points in the spatiotemporal association rule base are extracted, feature point combination processing is performed, and a key spatiotemporal feature set is generated.

5. The method according to claim 2, characterized in that: Based on the abnormal behavior trend prediction model, the federated transfer learning algorithm based on differential privacy is used to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate abnormal behavior urgency, including: Based on the abnormal behavior trend prediction model, performing interval distribution analysis on the risk score output by the abnormal behavior trend prediction model, evaluating the risk probability of different score intervals in the interval distribution analysis, and generating a risk probability distribution graph; Based on the risk probability distribution graph, a federated transfer learning algorithm based on differential privacy is used to perform an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, and generate a preliminary urgency grading result; Based on the preliminary urgency grading result, using causal inference technology, performing causal relationship analysis on the urgency grading analysis result, dynamically adjusting the grading threshold, and generating an optimized urgency grading result; Based on the optimized urgency grading result, a multi-level feedback mechanism is established, and different feedback paths are set for different levels of urgency in the optimized urgency grading result to generate abnormal behavior urgency.

6. The method according to claim 5, characterized in that Based on the risk probability distribution graph, the federated transfer learning algorithm based on differential privacy is used to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, extract key features of the risk score, dynamically set the grading threshold of the risk score, and generate a preliminary urgency grading result, including: Performing in-depth analysis based on the risk probability distribution graph, focusing on the overall risk pattern and trend in the risk probability distribution graph, vectorizing the overall risk pattern and trend, and generating a risk feature matrix; Based on the risk feature matrix, applying a federated transfer learning algorithm based on differential privacy to each risk feature vector in the risk feature matrix, performing an urgency grading analysis on the risk score output by the abnormal behavior trend prediction model, and generating an urgency grading score; Based on the urgency grading score, the key features of the risk score are extracted to perform score fusion processing, and according to the score fusion processing result, the grading threshold of the risk score is dynamically set to generate an optimized grading threshold; Based on the optimized classification threshold, an adaptive adjustment mechanism is introduced to record output data in real time during the operation of the abnormal behavior trend prediction model, and compare and analyze the optimized classification threshold to automatically adjust and generate a preliminary urgency classification result.

7. The method according to claim 1, characterized in that The customized safety notification is output based on the urgency of the abnormal behavior and sent to the target source group of the real-time environmental perception data, the user behavior and environmental changes of the target source group are continuously monitored, the abnormal pattern library is dynamically updated, and an intelligent identification and early warning strategy is generated, including: Outputting a customized safety notification based on the urgency of the abnormal behavior, and sending it to a target source group of the real-time environmental perception data, and generating a customized safety notification record; Based on the customized security notification record, an automated response process is set to automate the response process of the target source group after receiving the customized security notification record, and an automated response log is generated; Based on the automated response log, continuously monitor the user behavior and environmental changes of the target source group, capture potential abnormal activities during the continuous monitoring process, and generate a real-time monitoring report; Based on the real-time monitoring report, newly generated abnormal patterns in the real-time monitoring report are extracted to dynamically update the abnormal pattern library and generate intelligent identification and early warning strategies.

8. An intelligent identification and early warning system, characterized in that: include: A receiving module, used for receiving real-time environment perception data and predefining an abnormal pattern library, and performing dynamic matching processing on the real-time environment perception data based on the abnormal pattern library to generate a potential abnormal activity list; A processing module is used to dynamically capture the time series data and spatial correlation of abnormal activities in the abnormal activity list based on the potential abnormal activity list by using a spatiotemporal dynamic graph neural flow algorithm, and to perform weighted processing on key spatiotemporal feature points in the dynamic capture results by using a feature weighting method to generate an abnormal behavior trend prediction model; An analysis module is used to perform urgency grading analysis on the risk score output by the abnormal behavior trend prediction model based on the abnormal behavior trend prediction model by using a federated transfer learning algorithm based on differential privacy, extract key features of the risk score, dynamically set a grading threshold of the risk score, use causal inference technology to perform causal relationship analysis on the urgency grading analysis results, dynamically adjust the grading threshold, and generate abnormal behavior urgency; The monitoring module is used to output customized safety notifications based on the urgency of the abnormal behavior, send them to the target source group of the real-time environmental perception data, continuously monitor the user behavior and environmental changes of the target source group, dynamically update the abnormal pattern library, and generate intelligent recognition and early warning strategies.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an intelligent identification and early warning method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, an intelligent identification and early warning method as described in any one of claims 1 to 7 is implemented.

Images