Public security situation awareness and evaluation system driven by big data

Abstract

The invention relates to a big data-driven public security situation awareness and evaluation system and model, which can comprehensively integrate various information such as social media, surveillance videos, sensors, government databases and the like through a multi-source data dynamic weighted fusion algorithm, and adopts an improved hidden Markov model (IHMM) and the like to construct a public security situation awareness model. The dynamic change process of the public security event can be effectively processed, and the occurrence probability and the possible development state of the public security event can be accurately predicted through learning of the state transition probability matrix and the observation probability matrix. Compared with a traditional method based on experience or simple statistics, the method has higher accuracy and adaptability in the face of complex and changeable public security situations.

Description

Technical Field

[0001] The present invention relates to the field of public safety technology, and in particular to a public safety situation awareness and assessment system driven by big data. Background Art

[0002] In today's digital age, public safety management faces unprecedented challenges and opportunities. With the rapid development of information technology, various types of data are growing explosively, such as massive video data generated by urban surveillance cameras, relevant text and picture information posted by people on social media platforms, and environmental and safety parameter data collected by sensors distributed in various corners. However, traditional public safety situation awareness and assessment methods are mostly based on limited manual experience and simple statistical analysis, which are difficult to cope with such a large and complex data torrent. These methods often have problems such as difficulty in data integration, insufficient analysis depth, poor timeliness, and inability to effectively capture the complex correlation between multi-source data. As a result, in the early warning and response process of public safety incidents, it is impossible to grasp the situation changes in a timely and accurate manner, and it is difficult to formulate scientific and reasonable decision-making strategies, which seriously affects the efficiency and quality of public safety management. For example, in the security work of large-scale urban activities, traditional methods may miss the early identification and early warning of potential security risks due to the inability to integrate data from different regions, real-time feedback from people on social media, and environmental sensor data in a timely manner, which may lead to serious public safety incidents. Therefore, there is an urgent need for an innovative big data-driven public safety situation awareness and assessment system and model to break through these limitations and improve the level of intelligence in public safety management. Summary of the invention

[0003] The present invention provides a big data driven public safety situation awareness and assessment system and model, including: a data collection layer for collecting relevant data from social media platforms, various sensors and public safety databases;

[0005] The data preprocessing layer cleans, removes noise and unifies the format of the collected raw data;

[0006] The feature extraction and fusion layer uses a specific algorithm to extract key features from different types of data, and uses a multi-source data dynamic weighted fusion algorithm to fuse the features. In the multi-source data dynamic weighted fusion algorithm, assume that there are n data sources, and the feature vector of the i-th data source is X i =(x i1 ,x i2 ,…,x im ), with weight w i and The fused feature vector Y = (y1, y2, ..., y m ),in Weight w i Based on the correlation degree S between data source and public safety events i and data reliability s1 is determined, The situation awareness model layer builds a public security situation awareness model based on an improved hidden Markov model (IHMM), wherein the IHMM has a state set S = {s1, s2, ..., s k}、Observation set O={o1,o2,…,o l}, initial state probability distribution, state transition probability matrix A = {a ij} and the observation probability matrix B = {b j (o t )}

[0007] , the probability of public safety events and state sequences are calculated through the forward-backward algorithm; the situation assessment layer, based on the situation perception results, uses the fuzzy comprehensive evaluation method to conduct a comprehensive evaluation of the public safety situation. The fuzzy comprehensive evaluation method has an evaluation factor set U = {u1,u2,…,u p}、Evaluation level set V={v1,v2,…,v q}, by determining the single factor fuzzy evaluation matrix R = {Tij} and the factor weight vector W = (w1, w2, …, w p ), calculate the comprehensive evaluation vector Determine the severity level of the situation;

[0008] The user interaction layer is used to display situational awareness and assessment results, set system parameters, and perform data visualization analysis.

[0009] Furthermore, the social media data collected by the data collection layer extracts keywords, sentiment tendency, topic distribution and semantic similarity as feature vectors, wherein keyword weights are determined by a term frequency-inverse document frequency (TF-IDF) algorithm, and sentiment tendency is determined by a sentiment analysis model.

[0010] Furthermore, the data collected by the data acquisition layer extracts personnel behavior characteristics, target appearance characteristics, scene characteristics, personnel density information and target motion trajectory as feature vectors, personnel behavior characteristics and targets are identified through a deep learning target detection algorithm, and the target motion trajectory is obtained through a target tracking algorithm.

[0011] Furthermore, the sensor data collected by the data collection layer extracts environmental parameter characteristics and their change rates and fluctuation amplitudes as feature vectors, and the sensor data is calibrated and processed for outliers, and data interpolation is used to supplement missing data points.

[0012] Furthermore, in the fuzzy comprehensive evaluation method, the evaluation factor set includes at least one of social security conditions, fire safety conditions, traffic safety conditions and public health conditions.

[0013] Furthermore, in the fuzzy comprehensive evaluation method, the single-factor fuzzy evaluation matrix is ​​determined by expert evaluation, data statistical analysis or a combination of the two.

[0014] Furthermore, in the fuzzy comprehensive evaluation method, the factor weight vector is determined by using the hierarchical analysis method, which is obtained through the steps of constructing a judgment matrix, calculating a eigenvector and performing a consistency test.

[0015] Furthermore, in the multi-source data dynamic weighted fusion algorithm, the correlation between the data source and the public safety event is obtained by calculating the similarity between the data source characteristics and the historical public safety event characteristics, and the data reliability is determined by comprehensively considering the credibility of the data source, data integrity and accuracy.

[0016] Furthermore, the user interaction layer can adjust the parameters of the situation awareness model according to user instructions, and display the changes in perception and evaluation results after the parameters are adjusted.

[0017] Beneficial effects:

[0018] Through the dynamic weighted fusion algorithm of multi-source data, it can fully integrate various types of information such as social media and sensors, and use the improved hidden Markov model (IHMM) to build a public security situation awareness model, which can effectively handle the dynamic changes of public security events. Through the learning of state transition probability matrix and observation probability matrix, it can accurately predict the probability of occurrence and possible development status of public security events. Compared with traditional experience-based or simple statistical methods, it has higher accuracy and adaptability when facing complex and changing public security situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A structural block diagram of a big data-driven public security situation awareness and assessment system provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0020] Example 1

[0021] The data collection layer is used to collect data from social media platforms, various sensors, and public safety databases;

[0022] The data preprocessing layer cleans, removes noise and unifies the format of the collected raw data;

[0023] The feature extraction and fusion layer uses a specific algorithm to extract key features from different types of data, and uses a multi-source data dynamic weighted fusion algorithm to fuse the features. In the multi-source data dynamic weighted fusion algorithm, assume that there are n data sources, and the feature vector of the i-th data source is X i =(x i1 ,x i2 ,…,x im ), with weight w i and The fused feature vector Y = (y1, y2, ..., y m ),in Weight w i Based on the correlation degree S between data source and public safety events i and data reliability s1 is determined, The situation awareness model layer builds a public security situation awareness model based on an improved hidden Markov model (IHMM), wherein the IHMM has a state set S = {s1, s2, ..., s k}、Observation set O={o1,o2,…,o l}, initial state probability distribution, state transition probability matrix A = {a ij} and the observation probability matrix B = {b j (o t )}

[0024] , the probability of public safety events and state sequences are calculated through the forward-backward algorithm; the situation assessment layer, based on the situation perception results, uses the fuzzy comprehensive evaluation method to conduct a comprehensive evaluation of the public safety situation. The fuzzy comprehensive evaluation method has an evaluation factor set U = {u1,u2,…,u p}、Evaluation level set V={v1,v2,…,v q}, by determining the single factor fuzzy evaluation matrix R = {Tij} and the factor weight vector W = (w1, w2, …, w p ), calculate the comprehensive evaluation vector Determine the severity level of the situation;

[0025] The user interaction layer is used to display situational awareness and assessment results, set system parameters, and perform data visualization analysis.

[0026] The data collection layer extracts keywords, sentiment tendency, topic distribution and semantic similarity from the social media data collected as feature vectors, wherein keyword weights are determined by a term frequency-inverse document frequency (TF-IDF) algorithm, and sentiment tendency is determined using a sentiment analysis model.

[0027] The surveillance video data collected by the data acquisition layer extracts personnel behavior characteristics, target appearance characteristics, scene characteristics, personnel density information and target motion trajectory as feature vectors. Personnel behavior characteristics and targets are identified through a deep learning target detection algorithm, and the target motion trajectory is obtained through a target tracking algorithm.

[0028] The sensor data collected by the data acquisition layer extracts environmental parameter characteristics and their change rates and fluctuation amplitudes as feature vectors, and the sensor data is calibrated and processed for outliers, and data interpolation is used to supplement missing data points.

[0029] In the fuzzy comprehensive evaluation method, the evaluation factor set includes at least one of social security conditions, fire safety conditions, traffic safety conditions and public health conditions.

[0030] In the fuzzy comprehensive evaluation method, the single-factor fuzzy evaluation matrix is ​​determined by expert evaluation, data statistical analysis or a combination of the two.

[0031] In the fuzzy comprehensive evaluation method, the factor weight vector is determined by using the hierarchical analysis method, and is obtained through the steps of constructing a judgment matrix, calculating a eigenvector and performing a consistency test.

[0032] In the multi-source data dynamic weighted fusion algorithm, the correlation between the data source and the public safety event is obtained by calculating the similarity between the data source characteristics and the historical public safety event characteristics, and the data reliability is determined by comprehensively considering the credibility of the data source, data integrity and accuracy.

[0033] The user interaction layer can adjust the parameters of the situation awareness model according to user instructions and display the changes in perception and evaluation results after the parameters are adjusted.

[0034] The big data driven public security situation awareness and assessment method of the present invention comprises the following steps:

[0035] Step 1: Data collection: Collect public safety-related data from multiple sources through the data collection layer to ensure the real-time and comprehensiveness of the data.

[0036] Step 2: Data preprocessing: Clean, denoise, and unify the format of the collected raw data to remove invalid data and redundant information and improve data quality.

[0037] Step 3: Feature extraction and fusion. Use specific algorithms to extract key features from different types of data, such as extracting keywords, sentiment tendencies and other features from social media data, extracting personnel behavior, scene changes and other features from surveillance video data, and extracting environmental parameter features from sensor data. Then use a multi-source data dynamic weighted fusion algorithm to fuse these features and obtain a fused feature vector.

[0038] Step 4: Situation awareness. The fused feature vector is input into the public safety situation awareness model based on the improved hidden Markov model (IHMM), and the probability of occurrence and possible state sequence of public safety events are calculated through the forward-backward algorithm to realize the perception of the public safety situation.

[0039] Step 5: Situation assessment: Based on the situation awareness results, combined with expert experience and historical data, the fuzzy comprehensive evaluation method is used to conduct a comprehensive assessment of the public security situation and determine the severity level of the situation.

[0040] Step 6: User interaction. The user interaction layer displays situational awareness and assessment results to decision makers and relevant staff in an intuitive way, while providing functions such as system parameter settings and data visualization analysis to facilitate user decision-making and further data analysis.

[0041] Through the dynamic weighted fusion algorithm of multi-source data, various types of information such as social media, surveillance videos, sensors, and databases can be fully integrated, overcoming the limitations of traditional single data sources, making situation awareness information more comprehensive and accurate, and timely capturing potential public safety risks. For example, in the safety management of large-scale urban events, social media data can provide public sentiment and hot topic information, surveillance video data can capture on-site personnel behavior and scene changes, sensor data can monitor environmental anomalies, and databases provide statistical laws of historical events. The integration of this information can more accurately perceive the public safety situation. The public safety situation awareness model is constructed using the improved hidden Markov model (IHMM), which can effectively handle the dynamic changes of public safety events. Through the learning of the state transition probability matrix and the observation probability matrix, the probability of occurrence and possible development state of public safety events can be accurately predicted. Compared with traditional experience-based or simple statistical methods, it has higher accuracy and adaptability when facing complex and changing public safety situations. For example, when responding to an outbreak of an infectious disease, the IHMM model can accurately predict the development trend of the epidemic based on the dynamic update of multi-source data such as the number of new cases per day, changes in the epidemic spread area, and changes in people's prevention and control awareness and behavior, providing strong support for prevention and control decisions. The fuzzy comprehensive evaluation method comprehensively considers a variety of factors that affect the public security situation. It determines the evaluation factor set, evaluation level set and factor weight vector through expert experience and historical data, and can scientifically evaluate the severity level of the public security situation. This method avoids the one-sidedness of single indicator evaluation, provides decision makers with more comprehensive and objective evaluation results, and helps to formulate reasonable response strategies. For example, when evaluating the urban public security situation, comprehensive consideration of multiple factors such as social public security satisfaction is used to determine the risk level of the public security situation, providing a basis for the formulation of public security prevention and control measures. The multi-source data dynamic weighted fusion algorithm dynamically allocates weights according to the correlation between the data source and the public security incident and the reliability of the data, so that important data can play a greater role in the situation awareness process, avoid information bias caused by fixed weight allocation, and further improve the accuracy and reliability of situation awareness. At the same time, the IHMM model can continuously update the state transition probability matrix and observation probability matrix according to new data, adapt to the dynamic changes of the public security situation, and always maintain accurate perception of the situation. The user interaction layer provides decision makers and relevant staff with intuitive situation awareness and evaluation result display, as well as system parameter setting and data visualization analysis functions. Decision makers can adjust system parameters according to actual conditions and conduct in-depth data analysis to better understand the public security situation, formulate more scientific and reasonable response strategies, and optimize decision making. For example, when responding to emergencies, decision makers can quickly view relevant information about the event through the user interaction interface, adjust the parameters of the situation awareness model, evaluate the effects of different response strategies, and select the best response plan.Data collection equipment is fully deployed in three test sub-areas, including the surveillance camera network in the commercial center that fully covers key areas such as streets, shopping malls and parking lots, and collects topics and information related to the area on social media platforms, as well as deploying various environmental sensors to monitor parameters such as air quality and temperature; in the railway station area, in addition to surveillance cameras, it is also connected to the traffic management system to obtain train operation information, passenger flow data, etc., and uses sensors to monitor the safety of track facilities; in the industrial area, fire safety sensors, surveillance cameras, and connection with the internal safety management system of the enterprise to obtain worker operation information, etc. After preprocessing, feature extraction and fusion, the data is input into the situation awareness model based on the improved hidden Markov model (IHMM) and the situation assessment model of fuzzy comprehensive evaluation. The system outputs the public safety situation awareness results and assessment levels in real time, and displays them on the user interaction interface for reference by relevant departments. In area A, only the original surveillance cameras in the commercial center are used for video surveillance, and manual monitoring screens are regularly checked to judge abnormal behavior. In area B, the monitoring system and traffic flow counters at the railway station are mainly relied on. The flow of people and vehicles in the monitoring video is manually observed to determine whether there are safety hazards. The safety of rail facilities only relies on regular manual inspection records, lacking automated real-time monitoring and data fusion analysis. In area C, enterprises act independently, only simple fire safety alarms and surveillance cameras are set up in the factory, and data is not shared. The industrial zone management department only summarizes and analyzes the safety accident data reported by the enterprise, making it difficult to fully grasp the public safety situation of the entire area. In the comparative test, the public safety situation perception and evaluation system driven by big data of the present invention shows significant efficiency-enhancing advantages compared with the traditional public safety monitoring system. In terms of situation perception accuracy, the system of the present invention is more accurate and comprehensive in identifying event types. Traditional systems often miss a lot of key information due to a single data source and limited analysis methods, but the system of the present invention can accurately capture the characteristics of various events and accurately determine the event type. In terms of the location of the event, the traditional system relies on manual experience combined with limited monitoring images, and the error is large. However, the system of the present invention can greatly reduce the positioning error and accurately lock the location of the event by relying on multi-source data fusion and advanced algorithms. Traditional systems are almost unable to effectively predict the time of an event, but the system of the present invention can accurately estimate the time of an event based on dynamic changes in data and make preparations in advance. In terms of the effectiveness of situation assessment, the risk level assessment of the system of the present invention integrates multiple factors and has a scientific model. Compared with the traditional system's assessment method based on simple statistics and experience, the accuracy is greatly improved, which can provide a more reliable basis for decision-making. The recommended response strategies are also highly matched with actual needs due to accurate situation perception and scientific assessment. The response strategies of traditional systems often lack pertinence and timeliness, and it is difficult to effectively respond to complex public safety situations.In terms of response timeliness, the system of the present invention significantly shortens the time interval from the occurrence of an event to the issuance of an early warning and from the early warning to the relevant departments taking preliminary response measures. Traditional systems are far inferior to the system of the present invention in terms of response speed due to delayed information transmission and cumbersome manual processing procedures. This enables the system of the present invention to play a greater role in the prevention and handling of public safety incidents and effectively protect social public safety.

[0042] Example 2

[0043] A public safety incident early warning system based on big data and artificial intelligence, mainly covering data acquisition module, data integration and feature extraction module, intelligent analysis and prediction module, early warning decision module and user interaction and management module.

[0044] The data collection module collects a wide range of public safety-related data through various channels, such as using web crawler technology to capture real-time information from social media platforms, connecting to urban monitoring systems to obtain video data, connecting to various sensors to collect environmental and physical data, and downloading relevant statistical data from public safety databases.

[0045] The data integration and feature extraction module performs preprocessing operations such as cleaning, denoising, format conversion, and data alignment on the collected multi-source heterogeneous data, and then extracts key features of various types of data. For example, it extracts keywords, topic popularity, emotional tendencies and other features from social media texts; extracts personnel behavior features, object motion trajectory features, personnel density features, etc. from surveillance videos; extracts environmental parameter change features, equipment operation status features, etc. from sensor data; extracts event type distribution features, regional risk features, etc. from public security database data.

[0047] The intelligent analysis and prediction module adopts an innovative public safety incident warning model based on the fusion of deep belief network and support vector machine (DBN-SVM). First, the deep belief network (DBN) is used to perform unsupervised learning on the extracted multi-source data features, automatically extract high-level abstract features and potential patterns in the data, effectively reduce the data dimension and retain key information. DBN is composed of multiple restricted Boltzmann machines (RBMs) stacked together. It is trained by pre-training and reverse fine-tuning layer by layer, and can learn the complex nonlinear feature representation of the data. Then, the features learned by DBN are input into the support vector machine (SVM) for classification or regression prediction. SVM constructs the optimal classification hyperplane or regression function to achieve accurate prediction of the type, probability, time of occurrence and scope of impact of public safety events.

[0048] The early warning decision module formulates a response strategy based on the early warning results output by the intelligent analysis and prediction module, combined with the early warning decision model based on case reasoning and dynamic programming algorithms. Case-based reasoning searches the historical public safety event case library to find cases similar to the current early warning situation and draws on their successful response experience and strategies. The dynamic programming algorithm rationally allocates early warning resources and plans actions based on the current public safety resource status (such as police force distribution, rescue material reserves, medical resource allocation, etc.) and the dynamic changes in the development of the event, with the goal of minimizing losses and maximizing response effects, such as determining police force deployment plans, rescue action route planning, and personnel evacuation plans.

[0049] The user interaction and management module provides a convenient and intuitive operating platform for system administrators, decision makers and related staff to set system parameters, view early warning information, review and adjust response strategies, and visualize data, thereby realizing a human-machine collaborative public safety incident early warning management work mode.

[0050] On the other hand, the present invention also provides a public safety incident early warning method based on big data and artificial intelligence, comprising the following steps:

[0051] Step 1: Data collection. Use the data collection module to collect public safety-related data from multiple sources to ensure the integrity, timeliness and accuracy of the data, and provide sufficient data foundation for subsequent analysis. Step 2: Data preprocessing and feature extraction. Perform comprehensive preprocessing operations on the collected raw data to remove invalid data and noise interference, standardize and integrate data of different formats and types, and extract key features of various types of data so that subsequent models can effectively process and analyze data.

[0052] Step 3: Intelligent analysis and prediction based on DBN-SVM model. Construct a deep belief network (DBN) to perform unsupervised learning on the extracted multi-source data features. By training RBM layer by layer and fine-tuning the network parameters, high-level feature representation of the data is obtained. The features learned by DBN are input into the support vector machine (SVM). Based on the historical data and annotation information of public safety events, the SVM model is trained for classification or regression prediction, thereby obtaining relevant prediction information of public safety events, such as event type, probability of occurrence, time of occurrence, and scope of impact.

[0053] Step 4: Early warning decision making. According to the early warning results output by the intelligent analysis and prediction module, a response strategy is formulated using the early warning decision model based on case reasoning and dynamic programming algorithm. Based on case reasoning, similar historical cases are retrieved to obtain reference response strategies; the dynamic programming algorithm dynamically optimizes resource allocation and action planning based on current resources and events to determine the final early warning response plan.

[0054] Step 5: User interaction and system management. Through the user interaction and management module, relevant personnel can intuitively view warning information and response strategies, adjust system parameters according to actual conditions, manually intervene in and optimize warning decisions, and use data visualization tools to gain an in-depth understanding of public safety trends and system operation, thus achieving efficient warning management through human-machine collaboration.

[0055] Through the collaborative work of the data acquisition module and the data integration and feature extraction module, it is possible to widely collect and effectively integrate multi-source heterogeneous data from social media, surveillance videos, sensors, and databases, and use advanced feature extraction technology to mine key information in the data, providing rich and high-quality data resources for subsequent intelligent analysis and prediction, overcoming the problem of single data source and lack of information in traditional early warning systems. The early warning model based on the fusion of deep belief network and support vector machine (DBN-SVM) is adopted to give full play to the powerful ability of DBN in feature learning and data abstraction and the high-precision advantage of SVM in classification and regression prediction. DBN can automatically learn complex nonlinear features in data, effectively reduce data dimensions and retain key information, and provide SVM with better input features, thereby significantly improving the accuracy and reliability of public safety event prediction, especially when dealing with complex public safety events with multiple factors. The case-based reasoning and dynamic programming algorithms in the early warning decision module can fully draw on historical experience and combine the current actual situation to make dynamic optimization decisions. Case-based reasoning can quickly retrieve similar historical cases to provide preliminary response ideas and strategy references for current events, saving decision-making time and costs; the dynamic programming algorithm optimizes the allocation of warning resources and action plans based on the real-time status of public safety resources and the dynamic development trend of events with scientific and rigorous mathematical methods, ensuring that the warning response effect is maximized under limited resource conditions, effectively improving the efficiency and quality of public safety incident response. The user interaction and management module provides a friendly human-computer interaction interface, which facilitates managers and decision makers to participate in the entire process of public safety event warning. They can flexibly adjust system parameters based on actual experience and on-site conditions, manually review and intervene in warning information and decisions, and use data visualization to gain an in-depth understanding of the public safety situation and system operation status, realize the complementary advantages of man and machine, and further improve the adaptability and practicality of the system in complex and changeable public safety scenarios.

[0056] Example 3

[0057] A big data-driven public security situation awareness and assessment system mainly consists of a data acquisition subsystem, a data preprocessing and fusion subsystem, a situation awareness model construction and training subsystem, a situation assessment and early warning subsystem, and a user interaction and visualization subsystem.

[0058] The data collection subsystem comprehensively collects data related to public safety through a variety of advanced data collection technologies and extensive data source access. This includes but is not limited to: using distributed web crawler technology to capture real-time text, pictures, videos and other information from major social media platforms (such as Weibo, WeChat, Twitter, etc.), focusing on topics involving natural disasters, social security incidents, public health incidents and other public safety fields; deeply integrating with the city monitoring network to obtain high-resolution, multi-angle monitoring video stream data, covering key areas such as major streets, public places, and transportation hubs in the city; deploying a large-scale sensor network, covering environmental sensors (such as temperature, humidity, smoke, and harmful gas sensors), security sensors (such as infrared detectors, access control sensors, video motion sensors, etc.) and geographic information sensors (such as seismographs, water level sensors, etc.), real-time collection of environmental parameters, security status, and geographic spatial information data.

[0059] The data preprocessing and fusion subsystem performs preprocessing operations such as cleaning, denoising, format conversion, data alignment, and feature extraction on the collected massive multi-source heterogeneous data to improve data quality and convert it into a format suitable for model processing. Different preprocessing technologies are used for different types of data. For example, for social media text data, natural language processing technology is used to clean the text, remove irrelevant content such as advertisements, spam, emoticons, etc., perform part-of-speech tagging, named entity recognition and syntactic analysis, and extract text feature vectors such as keywords, topic popularity, and emotional tendency; for surveillance video data, video decoding, image enhancement, target detection and tracking, etc. are performed to extract video feature vectors such as personnel behavior characteristics, object motion trajectory characteristics, and personnel density characteristics; for sensor data, data calibration, outlier detection and elimination, data interpolation, etc. are performed to extract sensor feature vectors such as environmental parameter change characteristics and equipment operation status characteristics; for public safety database data, data cleaning, integration and conversion are performed to extract database feature vectors such as type distribution characteristics, regional risk characteristics, and time series characteristics of public safety events. After feature extraction, a fusion algorithm based on multimodal tensor decomposition is used to construct feature vectors of different modes into multimodal tensors. The potential correlation between different modal data is mined through tensor decomposition technology, and the deep fusion of multi-source data is realized to obtain the fused feature representation, which provides a rich and comprehensive data foundation for subsequent situation awareness and evaluation. The situation awareness model construction and training subsystem is based on an innovative hybrid neural network model - variational autoencoder and graph convolutional neural network fusion model (VAE-GCN) to build a public safety situation awareness model. First, the variational autoencoder (VAE) is used to encode and decode the fused feature data. VAE can effectively compress and reconstruct the data by learning the potential distribution characteristics of the data, extract the high-level abstract feature representation of the data, and introduce randomness in the encoding process to enhance the generalization and generation capabilities of the model. Then, the potential features encoded by VAE are used as the input of the graph convolutional neural network (GCN). GCN performs convolution operations based on graph structure data, which can fully mine the spatial topological relationship between data. In the public safety situation awareness scenario, different public safety related entities (such as personnel, places, events, etc.) are regarded as nodes in the graph, and edges are constructed according to the logical relationship between entities (such as the association between personnel and places, the occurrence relationship between events and places, etc.) to form a public safety situation graph. GCN performs convolution operations on the graph, propagates node information, and learns the mutual influence relationship between nodes, thereby realizing dynamic perception and modeling of public safety situation. In the model training process, a large-scale public safety historical data is combined with real-time data, and the model parameters are optimized through the back propagation algorithm, so that the model can accurately learn the evolution law and characteristic pattern of the public safety situation.

[0060] The situation assessment and early warning subsystem conducts a comprehensive assessment of the public security situation based on the results output by the situation awareness model, and generates early warning information based on the assessment results. The situation assessment model based on the fuzzy comprehensive evaluation method is used to determine the indicator system for public security situation assessment, including but not limited to indicators in multiple dimensions such as the possibility of event occurrence, the scope of event impact, the degree of event harm, and the heat of social public opinion. The evaluation results of multiple indicators are integrated into an overall situation assessment value through fuzzy transformation, and the assessment value can intuitively reflect the severity of the current public security situation. The evaluation value is compared with the pre-set early warning threshold, and the early warning mechanism is activated when the evaluation value exceeds the early warning threshold. The early warning information includes the type prediction of public security events, the time and location range of possible occurrence, the degree of harm estimation, and the corresponding response strategy recommendations, etc., providing timely and accurate decision-making basis for public security management departments. The user interaction and visualization subsystem provides an intuitive and convenient operation platform for public security management decision makers and relevant staff. Through this platform, users can view the results of public security situation awareness and assessment in real time, including visualized situation maps, trend charts, early warning information lists, etc., so as to intuitively understand the overall situation and changing trends of the public security situation. At the same time, users can also perform operations such as system parameter settings, data query and analysis, and early warning strategy adjustments on the platform to achieve human-computer interaction and collaborative work, and improve the flexibility and practicality of the system.

[0061] The data acquisition subsystem collects heterogeneous data from multiple sources, such as social media and sensor databases, and uses a fusion algorithm based on multimodal tensor decomposition to achieve deep data fusion. This can fully explore the potential correlation between different data sources and integrate various information resources, thereby achieving a comprehensive perception of the public safety situation. For example, in urban public safety management, the feedback from the public on social media about environmental anomalies in a certain area, the information on the gathering and abnormal behavior of people in the surveillance video of the area, the data on the increase in the concentration of harmful gases collected by environmental sensors, and the historical safety incident records of the area in the database are integrated and analyzed, which can more comprehensively and accurately grasp the public safety situation in the area and timely discover potential safety hazards. The situation awareness model is constructed by using the variational autoencoder and graph convolutional neural network fusion model (VAE-GCN), which gives full play to the advantages of VAE in data feature extraction and representation learning and the ability of GCN in processing graph structure data and mining spatial topological relationships. VAE can learn the potential distribution characteristics of data, extract high-level abstract features, and provide more representative input data for GCN; GCN effectively captures the mutual influence relationship between entities through convolution operations on the public safety situation map, and realizes dynamic perception and modeling of the situation. This fusion model can better adapt to the complexity and dynamics of public security data, improve the accuracy and reliability of situation awareness, and show stronger advantages in processing large-scale, multi-relational data. The situation assessment model based on the fuzzy comprehensive evaluation method can comprehensively consider the public security situation assessment indicators of multiple dimensions, and reasonably integrate the evaluation results of each indicator through fuzzy transformation, avoiding the one-sidedness and subjectivity of traditional single indicator evaluation or simple weighted average evaluation methods. The model can more scientifically and accurately reflect the overall situation and severity of the public security situation, and provide a more reliable basis for early warning decision-making. For example, when evaluating a possible public health event, considering the possibility of the event (such as calculated according to the epidemic transmission model), the scope of influence (such as the region and the number of people involved), the degree of harm (such as the degree of threat to public health and the impact on social economy) and the heat of social public opinion (such as the public's attention to the event and the degree of panic spread) and other indicators, it can comprehensively and accurately assess the event situation and formulate a more reasonable response strategy. The user interaction and visualization subsystem provides a friendly human-computer interaction interface, which is convenient for public security management decision makers and staff to participate in the system operation process. They can monitor the system in real time, adjust parameters and optimize early warning strategies based on actual experience and on-site conditions, and realize human-machine collaboration. For example, in special events or emergency situations, managers can make temporary adjustments to early warning thresholds based on their own judgment, or customize personalized situation visualization reports based on actual needs, thereby improving the adaptability and practicality of the system in complex and ever-changing public safety scenarios (the data information sources and uses involved in the present invention are legal and compliant and meet relevant management requirements).

Claims

1. A big data-driven public security situation awareness and assessment system, characterized in that: include: The data collection layer is used to collect relevant data from social media platforms, various sensors, and public safety databases; The data preprocessing layer cleans, removes noise and unifies the format of the collected raw data; The feature extraction and fusion layer uses a specific algorithm to extract key features from different types of data, and uses a multi-source data dynamic weighted fusion algorithm to fuse the key features. In the multi-source data dynamic weighted fusion algorithm, assume that there are n data sources, and the feature vector of the i-th data source is X i =(x i1 ,x i2 ,…,x im ), with weight w i and w i =1, the fused feature vector Y = (y1, y2, ..., y m ),in w i x ij , weight w i Based on the correlation degree S between data source and public safety events i and data reliability s1 is determined, The situation awareness model layer builds a public security situation awareness model based on an improved hidden Markov model (IHMM). The hidden Markov model has a state set S = {s1, s2, ..., s k }、Observation set O={o1,o2,…,o l }, initial state probability distribution, state transition probability matrix A = {a ij } and the observation probability matrix B = {b j (o t )}, the probability of public safety events and the state sequence are calculated by the forward-backward algorithm; the situation assessment layer, based on the situation perception results, uses the fuzzy comprehensive evaluation method to conduct a comprehensive evaluation of the public safety situation, and the fuzzy comprehensive evaluation method has an evaluation factor set U = {u1,u2,…,u p }、Evaluation level set V={v1,v2,…,v q }, by determining the single factor fuzzy evaluation matrix R = {Tij} and the factor weight vector W = (w1, w2, …, w p ), calculate the comprehensive evaluation vector Determine the severity level of the situation; The user interaction layer is used to display situational awareness and assessment results, set system parameters, and perform data visualization analysis.

2. The big data driven public security situation awareness and assessment system according to claim 1 is characterized in that: The data collection layer extracts keywords, sentiment tendencies, topic distribution and semantic similarity from the social media data collected as feature vectors, wherein keyword weights are determined by a word frequency-inverse document frequency algorithm, and sentiment tendencies are determined using a sentiment analysis model.

3. The big data driven public security situation awareness and assessment system according to claim 1 is characterized in that: The data collected by the data acquisition layer extracts personnel behavior characteristics, target appearance characteristics, scene characteristics, personnel density information and target motion trajectory as feature vectors. Personnel behavior characteristics and targets are identified through a deep learning target detection algorithm, and the target motion trajectory is obtained through a target tracking algorithm.

4. The big data driven public security situation awareness and assessment system according to claim 1, characterized in that: The sensor data collected by the data acquisition layer extracts environmental parameter characteristics and their change rates and fluctuation amplitudes as feature vectors, and the sensor data is calibrated and processed for outliers, and data interpolation is used to supplement missing data points.

5. The big data driven public security situation awareness and assessment system according to claim 1, characterized in that: In the fuzzy comprehensive evaluation method, the evaluation factor set includes at least one of social security conditions, fire safety conditions, traffic safety conditions and public health conditions.

6. The big data driven public security situation awareness and assessment system according to claim 1, characterized in that: In the fuzzy comprehensive evaluation method, the single-factor fuzzy evaluation matrix is ​​determined by expert evaluation, data statistical analysis or a combination of the two.

7. The big data driven public security situation awareness and assessment system according to claim 1, characterized in that: In the fuzzy comprehensive evaluation method, the factor weight vector is determined by using the hierarchical analysis method, and is obtained through the steps of constructing a judgment matrix, calculating a eigenvector and performing a consistency test.

8. The big data driven public security situation awareness and assessment system according to claim 1, characterized in that: In the multi-source data dynamic weighted fusion algorithm, the correlation between the data source and the public safety event is obtained by calculating the similarity between the data source characteristics and the historical public safety event characteristics, and the data reliability is determined by comprehensively considering the credibility of the data source, data integrity and accuracy.

9. The big data driven public security situation awareness and assessment system according to claim 1, characterized in that: The user interaction layer can adjust the parameters of the situation awareness model according to user instructions and display the changes in perception and evaluation results after the parameters are adjusted.

Images