An intelligent traffic management system based on space-time sequence features

Abstract

This invention discloses an intelligent traffic management system based on spatiotemporal sequence features, belonging to the field of intelligent traffic management technology. The system includes a data fusion and spatiotemporal sequence management module for real-time acquisition of multi-dimensional traffic data and data preprocessing; a spatiotemporal causal chain modeling and source analysis module for modeling spatiotemporal causal chains, assessing the causal strength of road events, and implementing causal control decisions; a traffic state prediction and intelligent control module for predicting traffic states and comprehensively analyzing traffic control needs; an anomaly detection and adaptive optimization module for assessing traffic anomalies and implementing adaptive traffic control measures; and an intelligent feedback and self-evolution module for comprehensively analyzing control benefit feedback and automatically implementing optimization strategies. This invention solves the problem of untimely traffic control in complex traffic environments where the superposition of multiple minor traffic anomalies makes it difficult to automatically perform intelligent source tracing of traffic event causal chains.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic management technology, specifically to an intelligent traffic management system based on spatiotemporal sequence characteristics. Background Technology

[0002] With the continuous growth of urban motor vehicle ownership and traffic demand, urban transportation networks are facing increasingly severe operational pressure. At the same time, the rapid development of traffic monitoring technology enables high-frequency and refined collection of multi-source data, such as geomagnetic data, video, floating car data, and sensor data, to monitor the operational status of the road network, providing a solid data foundation for intelligent traffic management and scientific decision-making.

[0003] For example, invention patent CN112598182B discloses an intelligent scheduling method and system for rail transit. The method includes: collecting raw data from multiple data sources within a preset range; obtaining passenger flow data of rail transit stations based on the raw data; constructing a passenger flow prediction model for the rail transit stations based on the passenger flow data; obtaining the passenger flow of the rail transit stations during a preset time period through the passenger flow prediction model; and coordinating the scheduling of vehicles and passengers at the rail transit stations based on the passenger flow during the preset time period to ensure smooth passenger flow at the rail transit stations. This invention obtains raw data from multiple data sources, and then obtains passenger flow data, achieving comprehensive monitoring of passenger flow at rail transit stations. It can issue different early warnings to enable reasonable coordination and scheduling of vehicles and passengers at rail transit stations, thereby orderly responding to upcoming large passenger flows and ensuring smooth passenger flow at rail transit stations.

[0004] For example, invention patent CN109685233B discloses a ground-air combined intelligent transportation system, including: an intelligent vehicle; a multi-level parking garage for storing and releasing the intelligent vehicle; a flight rotor configured to be detachably connected to the intelligent vehicle, and after the flight rotor is released, it reaches a designated location and connects with the intelligent vehicle to form a flying car capable of flight; and a rotor tower for storing and releasing the flight rotor. In this ground-air combined intelligent transportation system, users can book car travel and flight services through a smart terminal. The intelligent vehicle can automatically pick up and drop off passengers, and the flight rotor can automatically combine with the intelligent vehicle to deliver users to their destinations via an optimized route in a shorter time, reducing commuting time and saving travel costs.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0006] In complex traffic environments, existing traffic management methods struggle to identify and trace the causal chain of traffic incidents in a timely and automatic manner. This results in the inability to quickly pinpoint the source and propagation path of problems after accidents, congestion, or other anomalies occur, leading to delayed traffic control responses, inaccurate handling, and challenges that affect the overall operational efficiency and safety of the road network.

[0007] Therefore, in order to address the above problems, there is an urgent need for an intelligent traffic management system based on spatiotemporal sequence characteristics. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides an intelligent traffic management system based on spatiotemporal sequence features. This system solves the problem of untimely traffic control in complex traffic environments where the superposition of multiple minor traffic anomalies makes it difficult to automatically trace the causal chain of traffic events.

[0010] Technical solution

[0011] To achieve the above objectives, this invention provides the following technical solution: an intelligent traffic management system based on spatiotemporal sequence characteristics, comprising: a data fusion and spatiotemporal sequence management module, a spatiotemporal causal chain modeling and source analysis module, a traffic state prediction and intelligent control module, an anomaly detection and adaptive optimization module, and an intelligent feedback and self-evolution module; wherein, the data fusion and spatiotemporal sequence management module is used to collect multidimensional traffic data in real time and perform data preprocessing on the multidimensional traffic data; the spatiotemporal causal chain modeling and source analysis module is used to perform spatiotemporal causal chain modeling based on the preprocessed multidimensional traffic data, evaluate the causal strength of road events, and implement causal control decisions based on the causal strength evaluation results of road events; the traffic state prediction and intelligent control module... The traffic control module is used to predict traffic conditions based on real-time and historical preprocessed multidimensional traffic data, and to comprehensively analyze traffic control needs based on the traffic condition prediction results. The anomaly detection and adaptive optimization module is used to assess traffic anomalies based on the preprocessed multidimensional traffic data and traffic condition prediction results, and to implement adaptive traffic control measures by combining the traffic anomaly assessment results with the spatiotemporal causal chain, and to obtain the final multidimensional traffic data after control. The intelligent feedback and self-evolution module is used to obtain the final multidimensional traffic data after control and the traffic anomaly assessment results, and to analyze the control benefit feedback by combining the multidimensional traffic data and traffic anomaly assessment results before causal control and adaptive traffic control, and to automatically implement optimization strategies based on the analysis results.

[0012] Furthermore, the specific process of real-time acquisition and preprocessing of multi-dimensional traffic data is as follows: Multi-dimensional traffic data is collected in real time using various types of traffic sensors and urban traffic geographic information systems. This multi-dimensional traffic data includes: traffic flow, average vehicle speed, vehicle stationary time, road events, traffic light status, and road information. Traffic flow, average vehicle speed, and vehicle stationary time are compared with traffic flow thresholds, speed thresholds, and stationary time thresholds, respectively. Anomaly scores are obtained through a weighted normalization scoring method, and these anomaly scores are normalized to obtain road event intensity values. The multi-dimensional traffic data undergoes timestamp synchronization and spatial positioning, followed by missing value imputation, anomaly removal, noise reduction, and deduplication. The multi-dimensional traffic data is also normalized. Weights are assigned based on the importance of monitoring points, and the data acquisition frequency and priority are dynamically adjusted. A high-performance spatiotemporal database is established, and the multi-dimensional traffic data is stored in this database.

[0013] Furthermore, the specific process of spatiotemporal causal chain modeling based on multidimensional traffic data after data preprocessing is as follows: using the device identifier of traffic monitoring equipment as the identifier of graph nodes, establishing the adjacency relationship between nodes based on the actual road topology and traffic flow direction, constructing a dynamic spatiotemporal directed graph with traffic flow, average vehicle speed, and road events as the edges of the graph, and recording the road event timestamps and associated multidimensional traffic data of each monitoring node; based on the dynamic spatiotemporal directed graph, taking the road events collected by each monitoring node at continuous time as input, including accidents, congestion, and sudden changes in traffic flow, establishing a multivariate spatiotemporal event sequence dataset, training the spatiotemporal event sequence dataset using a greedy search algorithm, mining the probabilistic dependencies between road events under adjacent time and spatial adjacency relationships, constructing a dynamic Bayesian network model based on the probabilistic dependencies, and outputting the key causal nodes and chain reaction chains between road events.

[0014] Furthermore, the specific process of assessing the causal intensity of road events and implementing causal control decisions based on the assessment results is as follows: Obtain multi-dimensional traffic data; calculate the absolute traffic flow difference between node j at time t and node i at the previous time; obtain the road event intensity value between node i at the previous time and node j at time t; obtain the road event intensity value between node i and node j at time t; calculate the shortest road distance between nodes to obtain the actual road distance between node i and node j; multiply the road event intensity value between node i at the previous time and node j at time t, then divide by the sum of the actual road distance between node i and node j and a constant to obtain the total causal impact value; and then multiply the total causal impact value by the absolute traffic flow difference. The causal intensity assessment value of a road event is obtained by summing the traffic flow differences. The causal intensity assessment value is compared with the intensity threshold in real time. When the causal intensity assessment value is less than the intensity threshold, it is determined to be a general correlation, and routine traffic monitoring and automatic scheduling are adopted to continuously track changes in the road network status. When the causal intensity assessment value is greater than or equal to the intensity threshold, it is considered that there is a significant causal influence between nodes. The source is traced first and the upstream key nodes are located. Traffic light timing is dynamically optimized and emergency resources are linked. Real-time risk warnings and detour suggestions are pushed to drivers through vehicle terminals and navigation platforms. The propagation link and impact range of road events are displayed in real time, and the source and causal chain of abnormal events are automatically marked. A source tracing report is generated to clarify the initiation point, path and impact range of the anomaly.

[0015] Furthermore, the specific process of predicting traffic conditions based on real-time and historical preprocessed multidimensional traffic data is as follows: feature extraction and sample construction are performed on real-time and historical multidimensional traffic data and road event intensity values ​​as multidimensional spatiotemporal feature input datasets. The multidimensional spatiotemporal feature input datasets are trained using a long short-term memory network algorithm to construct a traffic condition prediction model, and output the prediction results of traffic flow, average vehicle speed, road events, and road event intensity values ​​at each traffic node at future times.

[0016] Furthermore, the specific process of comprehensively analyzing traffic control demand based on traffic state prediction results is as follows: obtain the predicted traffic flow of node i at future times from the traffic state prediction model output; obtain the road event intensity value of node i at future times as the predicted road event intensity value; obtain the predicted traffic flow of node i multiple times, and obtain the traffic flow prediction variance of node i at future times by calculating the sample variance; obtain the average vehicle speed of node i at future times as the average vehicle speed prediction value; multiply the predicted traffic flow, the sum of the predicted road event intensity value and constant one, and the traffic flow prediction variance to obtain the predicted queue fluctuation value; divide the predicted queue fluctuation value by the sum of the average vehicle speed prediction value and constant one to obtain the predicted control demand value. The system compares the predicted control demand value with the control threshold in real time. When the control demand value is less than the control threshold, it automatically optimizes the traffic light timing based on the basic traffic flow and prediction results. It continuously tracks the traffic status of nodes and continues to analyze the predicted control demand. It calculates the predicted control demand value three times in a row. If the predicted control demand value increases three times in a row, it switches to a key intervention response. When the control demand value is greater than or equal to the control threshold, it automatically extends the green light duration of nodes, optimizes the signal phase, and coordinates with surrounding intersections to carry out regional coordinated traffic management, prioritizing the relief of high-pressure road sections. It pushes real-time risk warnings, detour and diversion suggestions through vehicle navigation and road guidance screens. It records the intervention effect and outputs dynamic reports on high-pressure nodes and links.

[0017] Furthermore, based on the preprocessed multidimensional traffic data and traffic state prediction results, the specific process for traffic anomaly assessment is as follows: Obtain the traffic flow at node i at the current time t and the predicted traffic flow at future times; subtract the traffic flow at node i at the current time t from the predicted traffic flow at future times and take the absolute value to obtain the traffic flow change magnitude; obtain the average vehicle speed at node i at the current time t and the predicted average vehicle speed at future times; subtract the average vehicle speed at node i at the current time t from the predicted average vehicle speed at future times and take the absolute value to obtain the average vehicle speed change magnitude; obtain the road event intensity value at node i at the current time t; multiply the traffic flow change magnitude by the traffic flow weight factor to obtain the queue mutation term; multiply the average vehicle speed change magnitude by the speed weight factor to obtain the speed mutation term; multiply the road event intensity value at node i at the current time t by the event intensity weight factor to obtain the event intensity term; add the queue mutation term, speed mutation term, and event intensity term to obtain the comprehensive anomaly assessment value.

[0018] Furthermore, the specific process of implementing adaptive traffic control measures and obtaining the final multidimensional traffic data after combining traffic anomaly assessment results with spatiotemporal causal chains is as follows: Based on historical multidimensional traffic data and the output results of traffic state prediction models, dynamically set and adjust the hierarchical anomaly detection thresholds for each road segment, time period, and type; compare the anomaly comprehensive assessment value with the anomaly detection threshold in real time. When the anomaly comprehensive assessment value is less than the anomaly detection threshold, it is considered that there is no anomaly at the current moment, and routine traffic flow monitoring and multidimensional traffic data collection are maintained, while continuously monitoring traffic anomalies; when the anomaly comprehensive assessment value is greater than or equal to the anomaly detection threshold, it is considered that there is anomaly at the current moment. Combine the key causal nodes and chain reaction chains between road events and automatically backtrack along the traffic network to analyze whether the anomaly is local and independent, chain-like propagation, or large-scale diffusion, and assess the scope of impact and affected nodes accordingly; automatically optimize traffic light timing, extend the release time, push diversion, detour, and speed reduction suggestions to vehicle terminals and navigation platforms, and coordinate emergency resources for early response, while simultaneously updating the high-pressure node and link dynamic reports of the management platform.

[0019] Further, the specific process for analyzing the feedback of control benefits by obtaining the multidimensional traffic data and traffic anomaly assessment results after the final control, and combining them with the multidimensional traffic data and traffic anomaly assessment results before causal control and adaptive traffic control, is as follows: Collect multidimensional traffic data and comprehensive anomaly assessment values ​​after the optimization decision and store them in a high-performance spatiotemporal database; evaluate the control benefits by comparing the changes in multidimensional traffic data and comprehensive anomaly assessment values ​​before and after control for each node; calculate the comprehensive anomaly assessment values ​​of node i before and after control; obtain the magnitude of change in the comprehensive anomaly assessment value by subtracting the comprehensive anomaly assessment values ​​of node i before and after control and taking the absolute value; obtain the number of vehicle behavior changes as the amount of feedback data obtained by node i by subtracting the traffic flow of node i before and after control and taking the absolute value, and obtain the road event intensity value of node i after control; multiply the magnitude of change in the comprehensive anomaly assessment value by the amount of feedback data, and then divide by the sum of the road event intensity value of node i after control and a constant to obtain the control benefit feedback value of node i.

[0020] Furthermore, the specific process of automatically implementing optimization strategies based on the analysis results is as follows: For nodes with regulation benefit feedback values ​​greater than or equal to the feedback threshold, the optimization step size and sampling frequency are automatically adjusted, and samples with high regulation benefit feedback values ​​are used as key training samples for the traffic state prediction model. The traffic state prediction model, signal control strategy, and traffic anomaly assessment rules are automatically iterated and updated. Cases with high regulation benefit feedback values ​​are continuously accumulated and automatically archived, and optimization strategies and regulation effectiveness are implemented to build a knowledge and strategy base. For nodes with regulation benefit feedback values ​​less than the feedback threshold, the regular optimization frequency is maintained.

[0021] Beneficial effects

[0022] The present invention has the following beneficial effects:

[0023] (1) This invention models the spatiotemporal causal chain based on multidimensional traffic data after data preprocessing, evaluates the causal intensity of road events, and implements causal control decisions based on the causal intensity evaluation results of road events, thereby realizing the automatic identification and accurate tracing of key causal nodes and chain reaction chains between road events, thereby improving the scientificity and timeliness of abnormal event management.

[0024] (2) This invention collects multi-dimensional traffic data in real time, performs data preprocessing on the multi-dimensional traffic data, and establishes a high-performance spatiotemporal database to ensure the dynamic fusion, spatiotemporal synchronization and efficient storage of multi-source traffic information, laying a solid foundation for intelligent control of the entire process and subsequent data analysis.

[0025] (3) This invention uses multi-dimensional traffic data to predict traffic conditions and comprehensively analyzes traffic control needs based on the traffic condition prediction results. It can predict the traffic flow, average speed, road events and road event intensity values ​​of each traffic node at future times, providing support for the automatic generation of dynamic signal optimization and multi-objective control strategies.

[0026] (4) This invention assesses traffic anomalies based on multidimensional traffic data and traffic state prediction results after data preprocessing, and implements adaptive traffic control measures by combining the traffic anomaly assessment results with the spatiotemporal causal chain. It obtains the final multidimensional traffic data and traffic anomaly assessment results after control, and analyzes the control benefit feedback by combining the multidimensional traffic data and traffic anomaly assessment results before causal control and adaptive traffic control, thereby further implementing optimization strategies and promoting the continuous self-learning and intelligent evolution of the system.

[0027] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0028] Figure 1 This is a block diagram of an intelligent traffic management system based on spatiotemporal sequence features;

[0029] Figure 2 This is a directed network graph of causal strength for road events.

[0030] Figure 3 This is a time-series trend chart of the comprehensive evaluation value of node anomalies. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figures 1-3 This invention provides a technical solution: an intelligent traffic management system based on spatiotemporal sequence features, comprising: a data fusion and spatiotemporal sequence management module, a spatiotemporal causal chain modeling and source analysis module, a traffic state prediction and intelligent control module, an anomaly detection and adaptive optimization module, and an intelligent feedback and self-evolution module; wherein, the data fusion and spatiotemporal sequence management module is used to collect multidimensional traffic data in real time and perform data preprocessing on the multidimensional traffic data; the spatiotemporal causal chain modeling and source analysis module is used to perform spatiotemporal causal chain modeling based on the preprocessed multidimensional traffic data, evaluate the causal strength of road events, and implement causal control decisions based on the causal strength evaluation results of road events; the traffic state prediction and intelligent control module... The system is designed to predict traffic conditions based on real-time and historical preprocessed multidimensional traffic data, and to comprehensively analyze traffic control needs based on the traffic condition prediction results. An anomaly detection and adaptive optimization module is used to assess traffic anomalies based on the preprocessed multidimensional traffic data and traffic condition prediction results, and to implement adaptive traffic control measures by combining the traffic anomaly assessment results with the spatiotemporal causal chain, ultimately obtaining the final controlled multidimensional traffic data. An intelligent feedback and self-evolution module is used to acquire the final controlled multidimensional traffic data and traffic anomaly assessment results, and to analyze the control benefit feedback by combining the multidimensional traffic data and traffic anomaly assessment results before causal control and adaptive traffic control, and to automatically implement optimization strategies based on the analysis results.

[0033] Specifically, the real-time acquisition of multi-dimensional traffic data and the data preprocessing process are as follows: multi-dimensional traffic data is acquired in real time through various types of traffic sensors, including geomagnetic detectors, video surveillance cameras, microwave radar, infrared sensors, induction coils, radio frequency tags, floating car data acquisition terminals, and environmental meteorological sensors, and urban traffic road geographic information systems, which work together to ensure the wide coverage, high accuracy and real-time performance of the data. Multidimensional traffic data includes: traffic flow, average vehicle speed, vehicle stationary time, road events, traffic light status, and road information. Traffic flow, average vehicle speed, and vehicle stationary time are compared with traffic flow thresholds, speed thresholds, and stationary time thresholds, respectively. Anomaly scores are obtained through a weighted normalization scoring method. These anomaly scores are then normalized to obtain road event intensity values, which range from 0 to 1. The multidimensional traffic data undergoes timestamp synchronization and spatial positioning, followed by missing value imputation, anomaly removal, noise reduction, and deduplication to improve data quality. The data is also normalized. Weights are assigned to monitoring points based on their importance, automatically determining their current business focus level based on real-time traffic conditions, historically high-incidence anomaly areas, and traffic flow trends. For high-weight monitoring points, data collection frequency and priority are automatically and dynamically adjusted during emergencies and peak traffic periods. A high-performance spatiotemporal database is established, and the multidimensional traffic data is stored within it.

[0034] In this implementation plan, multi-dimensional traffic data is collected in real time through various types of traffic sensors and urban traffic road geographic information systems. Various traffic indicators are compared with thresholds, and anomaly scores and road event intensity values ​​are obtained through weighted normalization scoring. Combined with timestamp synchronization, spatial positioning, and intelligent adjustment of collection frequency, a high-performance spatiotemporal database is established, realizing efficient fusion of multi-dimensional traffic data and anomaly identification.

[0035] Specifically, the process of modeling spatiotemporal causal chains based on multidimensional traffic data after data preprocessing is as follows: using the device identifier of traffic monitoring equipment as the identifier of graph nodes, establishing adjacency relationships between nodes based on actual road topology and traffic flow direction, constructing a dynamic spatiotemporal directed graph with traffic flow, average vehicle speed, and road events as edges, and recording the road event timestamps and associated multidimensional traffic data of each monitoring node; based on the dynamic spatiotemporal directed graph, using road events collected by each monitoring node at continuous time intervals as input, including accidents, congestion, and sudden changes in traffic flow, establishing a multivariate spatiotemporal event sequence dataset, training the spatiotemporal event sequence dataset using a greedy search algorithm, and mining the probabilistic dependencies between road events under adjacent time and spatial adjacency relationships. The greedy search algorithm has the advantages of high computational efficiency and the ability to quickly converge to the optimal or suboptimal dependency structure in large-scale data scenarios, making it suitable for spatiotemporal feature mining of complex traffic networks. A dynamic Bayesian network model is constructed based on probabilistic dependencies, which outputs key causal nodes and chain reactions between road events. Key causal nodes refer to monitoring nodes that have a decisive influence on the occurrence and spread of abnormal events in the causal chain and can serve as the basis for control and traceability.

[0036] In this implementation scheme, a dynamic spatiotemporal directed graph is constructed, and a greedy search algorithm is used to mine probabilistic dependencies in multivariate spatiotemporal event sequences. A dynamic Bayesian network model is then built to output key causal nodes and chain reaction links between road events, thereby achieving efficient modeling and accurate tracing of causal relationships in traffic events.

[0037] Specifically, the process of assessing the causal intensity of road events and implementing causal control decisions based on the assessment results is as follows: Obtain multidimensional traffic data; calculate the absolute traffic flow difference between node j at time t and node i at the previous time; obtain the road event intensity values ​​of node i at the previous time and node j at time t; obtain road information and calculate the shortest road distance between nodes to obtain the actual road distance between node i and node j; multiply the road event intensity values ​​of node i at the previous time and node j at time t, then divide by the sum of the actual road distance between node i and node j and a constant to obtain the total causal impact value; and then multiply the total causal impact value by the absolute traffic flow... The difference between the measured values ​​is added to obtain the causal intensity assessment value of the road event; the causal intensity assessment value of the road event is compared with the intensity threshold in real time. When the causal intensity assessment value of the road event is less than the intensity threshold, it is determined to be a general correlation, and routine traffic monitoring and automatic scheduling are adopted to continuously track changes in the road network status; when the causal intensity assessment value of the road event is greater than or equal to the intensity threshold, it is considered that there is a significant causal influence between nodes, and the upstream key nodes are prioritized for tracing and locating, traffic light timing is dynamically optimized, and emergency resources are linked. Real-time risk warnings and detour suggestions are pushed to drivers through vehicle terminals and navigation platforms; the propagation link and impact range of the road event are displayed in real time, and the source of the event and its causal chain are automatically marked for abnormal events; a source tracing report is generated to clarify the initiation point, path and impact range of the anomaly.

[0038] The specific formula for assessing the causal intensity of road incidents is as follows:

[0039] ;

[0040] In the formula, This represents the causal strength assessment value of road events from node i to node j, used to reflect the causal strength of the influence of upstream node i on downstream node j within a certain time period in the traffic network. The larger the value, the more directly the recent traffic changes and events at node i will affect and trigger events at node j. This represents the traffic flow at node j at time t and the traffic flow at node i at the previous time. The absolute difference in traffic flow reflects the abrupt changes and transmission of traffic flow between upstream and downstream nodes; Indicates that node i at time... The road event intensity value; Indicates that node j at time... The road event intensity value; the value of the road event intensity value ranges from 0 to 1; This is used to measure the impact of the road event intensity value of upstream node i at the previous moment on the event intensity of downstream node j at the current moment; This represents the actual road distance between node i and node j, reflecting the spatial damping effect.

[0041] Based on different nodes and their corresponding downstream nodes, and using the actual road distance to each node, traffic flow at adjacent times, and road event intensity value, the final road event causality intensity assessment value is calculated. Table 1 shows the road event causality intensity assessment value data.

[0042] Table 1. Data Table of Causal Intensity Assessment Values ​​for Road Incidents

[0043]

[0044] like Figure 2 As shown, this is a directed network diagram of road event causality strength provided in an embodiment of this application. Arrows in the diagram represent causal directions, and the brighter the arrow, the greater the causal strength. This diagram transforms the abstract values ​​in the data table into an intuitive node and link structure, allowing for the visual identification of high-impact links, main channels for anomaly propagation, anomaly propagation paths, and bottleneck sections in the network. (Based on Table 1 and...) Figure 2 It can be seen that links with higher causal intensity assessment values ​​for road events indicate that these nodes have a more significant impact on traffic events and abnormal transmission between them, and are key points for tracing the source of road events.

[0045] In this implementation plan, by acquiring multi-dimensional traffic data and combining absolute traffic flow difference, road event intensity value and actual road distance, the causal intensity assessment value of road events is compared with the intensity threshold in real time. This enables regular traffic monitoring and automatic scheduling for general correlations, as well as priority source tracing, dynamic optimization of traffic light timing, linkage of emergency resources, real-time display of propagation links and generation of source tracing reports under significant causal influences, effectively improving the pertinence and intelligence level of control response.

[0046] Specifically, the process of predicting traffic conditions based on real-time and historical preprocessed multidimensional traffic data is as follows: feature extraction and sample construction are performed on real-time and historical multidimensional traffic data and road event intensity values ​​as multidimensional spatiotemporal feature input datasets. The multidimensional spatiotemporal feature input datasets are trained using a long short-term memory network algorithm to construct a traffic condition prediction model. Deep learning is used to automatically learn and represent the temporal patterns and spatial correlations of traffic flow, improving the ability to capture complex traffic changes. The prediction results of traffic flow, average speed, road events, and road event intensity values ​​at each traffic node at future times are output.

[0047] In this implementation plan, feature extraction and sample construction are performed on real-time and historical multidimensional traffic data and road event intensity values. The multidimensional spatiotemporal feature input dataset is trained using a long short-term memory network algorithm to construct a traffic state prediction model. The model outputs the predicted traffic flow, average speed, road events, and road event intensity values ​​for each traffic node at future times, providing a forward-looking decision-making basis for traffic management and control.

[0048] Specifically, the process of comprehensively analyzing traffic control demand based on traffic state prediction results is as follows: Obtain the predicted traffic flow of node i at future times from the traffic state prediction model output; obtain the road event intensity value of node i at future times; obtain the predicted road event intensity value; obtain the predicted traffic flow of node i multiple times, and calculate the sample variance to obtain the predicted traffic flow variance of node i at future times; obtain the average vehicle speed of node i at future times as the predicted average vehicle speed value; comprehensively characterize the traffic pressure and fluctuation trend of the node; multiply the predicted traffic flow, the sum of the predicted road event intensity value and a constant, and the traffic flow prediction variance to obtain the predicted queue fluctuation value; divide the predicted queue fluctuation value by the sum of the predicted average vehicle speed value and a constant to obtain the predicted control demand value; compare the predicted control demand value with the control threshold in real time. When the demand value for traffic control is less than the control threshold, the system automatically optimizes traffic light timing based on the baseline traffic flow and forecast results, continuously tracks the traffic status of nodes, and continues to analyze the predicted demand for traffic control. It calculates the predicted demand value three times consecutively. If the three predicted demand values ​​increase consecutively, indicating that the later calculated demand value is greater than the previous calculated demand value, the system switches to a priority intervention response. When the demand value is greater than or equal to the control threshold, the system automatically extends the green light duration at nodes, optimizes the signal phase, and coordinates with surrounding intersections for regional traffic management, prioritizing the relief of high-pressure road sections. It also pushes real-time risk warnings, detour and diversion suggestions through in-vehicle navigation and roadside guidance screens. The system records the intervention effect and outputs dynamic reports on high-pressure nodes and links, achieving proactive, tiered, and collaborative intelligent traffic control.

[0049] The specific formula for predicting and regulating demand is as follows:

[0050] ;

[0051] In the formula, This represents the predicted control demand value of node i at time t, which is used to quantify the urgency of signal control and management recommendations under the future traffic conditions of a certain node. The higher the predicted control demand value, the greater the risk of congestion, anomalies and fluctuations in the next prediction period for that node. Indicates node i at a future time. The predicted traffic flow measures the expected congestion level of node i over a future period, reflecting the scale of vehicle backlog. The larger the predicted traffic flow, the more severe the queuing will occur at that node in the future. Indicates node i at a future time. The predicted road event intensity value reflects the risk level of abnormal events occurring at node i in the future; Indicates node i at a future time. The traffic flow forecast variance measures the uncertainty and volatility of future traffic flow forecasts. The larger the forecast variance, the more unstable and drastic the future traffic conditions. This represents the predicted queuing volatility value, used to comprehensively measure the queuing risk and volatility intensity of node i in the future period; Indicates node i at a future time. The predicted average vehicle speed reflects the smoothness of traffic flow at node i in the future. The larger the predicted average vehicle speed, the smoother the traffic, and the less congestion and control pressure there is.

[0052] In this implementation plan, by acquiring the predicted traffic flow, predicted road event intensity, traffic flow prediction variance, and average speed prediction values ​​output by the traffic state prediction model, the predicted control demand value is calculated, and the predicted control demand value is compared with the control threshold in real time. This enables automatic optimization of traffic light timing, key intervention response, regional coordinated traffic management, and risk alerts, thereby achieving hierarchical and coordinated intelligent traffic control.

[0053] Specifically, the process of traffic anomaly assessment based on preprocessed multidimensional traffic data and traffic state prediction results is as follows: Obtain the traffic flow at node i at the current time t and the predicted traffic flow at future times; calculate the absolute value of the difference between the traffic flow at node i at the current time t and the predicted traffic flow at future times to obtain the traffic flow change magnitude, reflecting the potential queuing and congestion trend at the node in the future; obtain the average vehicle speed at node i at the current time t and the predicted average vehicle speed at future times; calculate the absolute value of the difference between the average vehicle speed at node i at the current time t and the predicted average vehicle speed at future times to obtain the average vehicle speed change magnitude, highlighting the impact of speed fluctuations on anomaly risks; obtain the road event intensity value at node i at the current time t; multiply the traffic flow change magnitude by the traffic flow weight factor to obtain the queue mutation term; multiply the average vehicle speed change magnitude by the speed weight factor to obtain the speed mutation term; multiply the road event intensity value at node i at the current time t by the event intensity weight factor to obtain the event intensity term; add the queue mutation term, speed mutation term, and event intensity term to obtain the comprehensive anomaly assessment value.

[0054] The specific formula for the comprehensive assessment value of anomalies is as follows:

[0055] ;

[0056] In the formula, This represents the comprehensive anomaly assessment value of node i at time t, which is used to quantify the degree of traffic anomaly at the current node i. The larger the comprehensive anomaly assessment value, the more significant the changes in traffic flow and speed, and the more severe the abnormal events that have occurred at node i in the current time period. This represents the traffic flow at node i at time t. Indicates node i at a future time. Predicted traffic flow; The magnitude of traffic flow change is used to measure the sudden change in traffic flow at node i between the current and the predicted future time period. The greater the change, the more drastic the fluctuation in traffic conditions and the higher the probability of anomalies. This represents the average vehicle speed of node i at the current time t, which is obtained by acquiring the average vehicle speed. Indicates node i at a future time. The predicted average vehicle speed; This represents the magnitude of vehicle speed change, measuring the sudden change in vehicle speed at node i between the current and future time periods; This represents the road event intensity value of node i at time t, used to determine whether the node has any identified abnormal road events and the degree of abnormality. The traffic flow weighting factor is obtained by statistically analyzing the contribution of historical traffic flow changes to the comprehensive evaluation value of anomalies and fitting it using a linear regression algorithm. The value ranges from 0 to 1. The speed weighting factor is obtained by statistically analyzing the contribution of historical vehicle speed changes to the comprehensive evaluation value of anomalies and fitting it using a linear regression algorithm. Its value ranges from 0 to 1. The event intensity weighting factor is obtained by statistically analyzing the actual impact of historical road event intensity values ​​on the final comprehensive anomaly assessment value, using the actual impact as a label, and fitting the data using the least squares method. The value ranges from 0 to 1.

[0057] The traffic flow weight factor was set to 0.6, the speed weight factor to 0.3, and the event intensity weight factor to 0.1. With these three weight factors unchanged, the anomaly comprehensive evaluation value of the node was calculated over time based on different traffic flow, average vehicle speed, and road event intensity values. Table 2 shows the anomaly comprehensive evaluation value data.

[0058] Table 2. Comprehensive Assessment Values ​​of Anomalies

[0059]

[0060] like Figure 3 As shown, the time-series trend chart of the comprehensive evaluation value of node anomalies provided in the embodiments of this application demonstrates the dynamic changes in node anomaly risk; according to Table 2 and Figure 3It can be seen that, with the same weighting factors, the trend of the anomaly comprehensive evaluation value changes when the traffic flow, average speed and road event intensity values ​​of the node are different at different times.

[0061] In this implementation plan, by acquiring traffic flow, predicted traffic flow, average vehicle speed, predicted average vehicle speed, and road event intensity values, the plan calculates queue mutation term, speed mutation term, and event intensity term, and adds the three terms together to obtain an anomaly comprehensive evaluation value, thereby achieving the coordinated quantification of multi-source anomaly factors and providing a basis for intelligent decision-making and risk classification.

[0062] Specifically, the process of implementing adaptive traffic control measures by combining traffic anomaly assessment results with spatiotemporal causal chains, and obtaining the final multidimensional traffic data after control, is as follows: Based on historical multidimensional traffic data and the output of the traffic state prediction model, the hierarchical anomaly detection thresholds for each road segment, time period, and type are dynamically set and adjusted; the setting of the hierarchical anomaly detection thresholds includes the historical anomaly distribution of different road segments, traffic flow peak and valley characteristics, and the probability of anomaly events, and the levels are divided into normal and obviously abnormal according to the comprehensive anomaly assessment value; the comprehensive anomaly assessment value is compared with the anomaly detection threshold in real time, and when the comprehensive anomaly assessment value is less than the anomaly detection threshold, the traffic node is considered to be abnormal. If there are no anomalies at the current moment, maintain regular traffic flow monitoring and multi-dimensional traffic data collection, and continuously monitor traffic anomalies; when the comprehensive anomaly assessment value is greater than or equal to the anomaly detection threshold, it is considered that there is an anomaly at the current moment. Combine the key causal nodes and chain reaction chains between road events and automatically backtrack along the traffic network to analyze whether the anomaly is local and independent, chain-like propagation, or large-scale spread, and assess the scope of impact and affected nodes accordingly; automatically optimize traffic light timing, extend the green time, push diversion, detour, and speed reduction suggestions to vehicles and navigation platforms, and coordinate emergency resources to respond in advance, and simultaneously update the high-pressure node and link dynamic reports of the management platform.

[0063] In this implementation plan, based on historical multidimensional traffic data and the output of traffic condition prediction models, the hierarchical anomaly detection thresholds for each road segment, time period, and type are dynamically set and adjusted. Combined with the hierarchical response of the comprehensive anomaly assessment value, the plan enables routine traffic monitoring and data collection when there are no anomalies, automatic backtracking along the traffic network when there are anomalies, and implementation of anomaly control measures, thereby improving the hierarchical accuracy and efficiency of adaptive traffic control.

[0064] Specifically, the process of obtaining multidimensional traffic data and traffic anomaly assessment results after final regulation, and combining them with multidimensional traffic data and traffic anomaly assessment results before causal regulation and adaptive traffic regulation, to analyze the regulation benefit feedback is as follows: Collect multidimensional traffic data and comprehensive anomaly assessment values ​​after optimization decisions and store them in a high-performance spatiotemporal database; evaluate the regulation benefits by comparing the changes in multidimensional traffic data and comprehensive anomaly assessment values ​​before and after regulation at each node; comprehensively reflect the actual effect of optimization measures on traffic condition improvement; calculate the comprehensive anomaly assessment values ​​of node i before and after regulation; obtain the magnitude of change in the comprehensive anomaly assessment value by subtracting the comprehensive anomaly assessment values ​​of node i before and after regulation and taking the absolute value, to quantify the anomaly mitigation effect; obtain the number of vehicle behavior changes by subtracting the traffic flow of node i before and after regulation and taking the absolute value, as the amount of feedback data obtained by node i, and simultaneously obtain the road event intensity value of node i after regulation; multiply the magnitude of change in the comprehensive anomaly assessment value by the amount of feedback data, and then divide by the sum of the road event intensity value of node i after regulation and a constant to obtain the regulation benefit feedback value of node i.

[0065] The specific formula for the feedback value of the regulation return is as follows:

[0066] ;

[0067] In the formula, This represents the feedback value of the regulation benefit of node i, which is used to allocate more attention and learning resources to nodes with effective measures, sufficient feedback, and low risk, so as to achieve efficient self-evolution and continuous improvement of the level of intelligent traffic management; This represents the comprehensive abnormal assessment value of node i before regulation; This represents the comprehensive evaluation value of the anomaly at node i after regulation; This indicates the magnitude of change in the comprehensive assessment value for anomalies, reflecting the actual benefits and extent of improvement brought about by control measures; This represents the amount of feedback data obtained by node i. When traffic control is implemented, vehicles receive risk warnings and perform diversion or detour operations. By obtaining the actual reduction in traffic flow at node i before and after the control, the actual number of vehicles taking new routes is reflected, which further reflects how much effective feedback was obtained after the traffic control was implemented. This represents the road event intensity value of node i after regulation.

[0068] In this implementation plan, multidimensional traffic data and anomaly comprehensive evaluation values ​​after optimization decisions are collected and stored in a high-performance spatiotemporal database. The changes in multidimensional traffic data and anomaly comprehensive evaluation values ​​of nodes before and after regulation are compared, the magnitude of change in anomaly comprehensive evaluation values ​​and the number of changes in vehicle behavior are calculated, and combined with the road event intensity values ​​of nodes after regulation, the regulation benefit feedback value is obtained, thereby realizing quantitative evaluation and feedback-driven adaptive optimization.

[0069] Specifically, the process of automatically implementing optimization strategies based on the analysis results is as follows: For nodes whose regulation benefit feedback value is greater than or equal to the feedback threshold, the optimization step size and sampling frequency are automatically adjusted according to the regulation benefit feedback value. The optimization step size refers to the magnitude of each update when adjusting the traffic state prediction model parameters and signal timing. Increasing the step size can accelerate adaptation to changes, while decreasing the step size is smoother and more robust. The sampling frequency refers to the time interval between collecting and processing node data. A high sampling frequency can improve the sensitivity of anomaly detection, while a low sampling frequency saves resources. The parameter update speed and data collection density of nodes whose regulation benefit feedback value is greater than or equal to the feedback threshold are increased. Samples with high regulation benefit feedback values ​​are used as key training samples for the traffic state prediction model, and the traffic state prediction model, signal control strategy, and traffic anomaly assessment rules are automatically updated iteratively. Cases with high regulation benefit feedback values ​​are continuously accumulated and automatically archived to implement optimization strategies and regulation effectiveness, and to build a knowledge and strategy base. For nodes whose regulation benefit feedback value is less than the feedback threshold, the regular optimization frequency is maintained.

[0070] In this implementation plan, the traffic state prediction model and strategy are updated by automatically adjusting the optimization step size and sampling frequency, archiving cases with high regulation benefit feedback values, building a knowledge and strategy base, and implementing differentiated optimization frequencies for different feedback value nodes, so as to achieve continuous optimization and intelligent upgrading of the system.

[0071] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0072] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent traffic management system based on spatiotemporal sequence features, characterized in that, include: The modules include: data fusion and spatiotemporal sequence management, spatiotemporal causal chain modeling and source analysis, traffic state prediction and intelligent control, anomaly detection and adaptive optimization, and intelligent feedback and self-evolution. The data fusion and spatiotemporal sequence management module is used to collect multi-dimensional traffic data in real time and perform data preprocessing on the multi-dimensional traffic data. The spatiotemporal causal chain modeling and source tracing analysis module is used to perform spatiotemporal causal chain modeling based on multidimensional traffic data after data preprocessing, evaluate the causal intensity of road events, and implement causal control decisions based on the causal intensity evaluation results of road events. The traffic condition prediction and intelligent control module is used to predict traffic conditions based on real-time and historical preprocessed multi-dimensional traffic data, and to comprehensively analyze traffic control needs based on the traffic condition prediction results. The anomaly detection and adaptive optimization module is used to assess traffic anomalies based on the preprocessed multidimensional traffic data and traffic state prediction results, and to implement adaptive traffic control measures by combining the traffic anomaly assessment results with the spatiotemporal causal chain, and to obtain the final controlled multidimensional traffic data. The intelligent feedback and self-evolution module is used to acquire multi-dimensional traffic data and traffic anomaly assessment results after final regulation, and to analyze the regulation benefit feedback by combining the multi-dimensional traffic data and traffic anomaly assessment results before causal regulation and adaptive traffic regulation, and to automatically implement optimization strategies based on the analysis results. The specific process of obtaining the multidimensional traffic data and traffic anomaly assessment results after the final regulation, and combining them with the multidimensional traffic data and traffic anomaly assessment results before causal regulation and adaptive traffic regulation, to analyze the regulation benefit feedback is as follows: Collect multidimensional traffic data and anomaly comprehensive evaluation values ​​after optimized decision-making, and store them in a high-performance spatiotemporal database; evaluate the effectiveness of regulation by comparing the changes in multidimensional traffic data and anomaly comprehensive evaluation values ​​before and after regulation at each node; The abnormal comprehensive evaluation value of node i before and after regulation is calculated respectively; the difference between the abnormal comprehensive evaluation value of node i before and after regulation is taken as the absolute value to obtain the change range of the abnormal comprehensive evaluation value; the difference between the traffic flow of node i before and after regulation is taken as the absolute value to obtain the number of vehicle behavior changes as the amount of feedback data obtained by node i, and the road event intensity value of node i after regulation is obtained. Multiply the magnitude of the change in the comprehensive assessment value of the anomaly by the amount of feedback data, and then divide by the sum of the road event intensity value of node i after regulation and the constant one to obtain the regulation benefit feedback value of node i. The specific process of automatically implementing optimization strategies based on the analysis results is as follows: For nodes whose regulation benefit feedback value is greater than or equal to the feedback threshold, the optimization step size and sampling frequency are automatically adjusted based on the regulation benefit feedback value. The optimization step size refers to the magnitude of each update of the traffic state prediction model parameters and signal timing adjustments, while the sampling frequency refers to the time interval between collecting and processing node data. This improves the parameter update speed and data collection density for nodes whose regulation benefit feedback value is greater than or equal to the feedback threshold. Samples with high regulation benefit feedback values ​​are used as key training samples for the traffic state prediction model, automatically iteratively updating the traffic state prediction model, signal control strategies, and traffic anomaly assessment rules. Cases with high regulation benefit feedback values ​​are continuously accumulated and automatically archived to implement optimization strategies and regulation effectiveness, building a knowledge and strategy base. For nodes whose regulation benefit feedback value is less than the feedback threshold, the regular optimization frequency is maintained.

2. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process of real-time acquisition of multi-dimensional traffic data and data preprocessing of the multi-dimensional traffic data is as follows: Multidimensional traffic data is collected in real time using various types of traffic sensors and urban traffic road geographic information systems. The multidimensional traffic data includes: traffic flow, average vehicle speed, vehicle stationary time, road events, traffic light status, and road information. Traffic flow, average vehicle speed, and vehicle stationary time are compared with traffic flow thresholds, speed thresholds, and stationary time thresholds, respectively. Anomaly scores are obtained through a weighted normalization scoring method. The anomaly scores are then normalized to obtain the road event intensity value. The system performs time-stamp synchronization and spatial positioning on multidimensional traffic data, and performs missing value imputation, anomaly removal, noise reduction and deduplication operations. It also normalizes the multidimensional traffic data, sets weights based on the importance of monitoring points, and dynamically adjusts the data collection frequency and priority. A high-performance spatiotemporal database is established and the multidimensional traffic data is stored in the high-performance spatiotemporal database.

3. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process of performing spatiotemporal causal chain modeling based on preprocessed multidimensional traffic data is as follows: Using the device identifier of traffic monitoring equipment as the identifier of graph nodes, the adjacency relationship between nodes is established based on the actual road topology and traffic flow direction. With traffic flow, average vehicle speed, and road events as the edges of the graph, a dynamic spatiotemporal directed graph is constructed, and the road event timestamps and associated multidimensional traffic data of each monitoring node are recorded. Based on a dynamic spatiotemporal directed graph, road events collected by each monitoring node at continuous time intervals are used as input. Road events include accidents, congestion, and sudden changes in traffic flow. A multivariate spatiotemporal event sequence dataset is established. The spatiotemporal event sequence dataset is trained using a greedy search algorithm to mine the probabilistic dependencies between road events under adjacent temporal and spatial adjacency. Based on the probabilistic dependencies, a dynamic Bayesian network model is constructed to output the key causal nodes and chain reaction chains between road events.

4. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process of assessing the causal intensity of road events and implementing causal control decisions based on the assessment results is as follows: Acquire multidimensional traffic data, subtract the traffic flow of node j at time t from the traffic flow of node i at the previous time and take the absolute value to obtain the absolute traffic flow difference; obtain the road event intensity value of node i at the previous time and the road event intensity value of node j at time t; obtain road information and calculate the shortest road distance between nodes to obtain the actual road distance between node i and node j. Multiply the road event intensity value of node i at the previous time step by the road event intensity value of node j at time t, and then divide by the sum of the actual road distance between node i and node j and a constant to obtain the total causal impact value. Add the total causal impact value to the difference in absolute traffic flow to obtain the road event causal intensity assessment value. The causal intensity assessment value of a road event is compared with the intensity threshold in real time. When the causal intensity assessment value of a road event is less than the intensity threshold, it is determined to be a general association, and routine traffic monitoring and automatic scheduling are adopted to continuously track changes in the road network status. When the causal intensity assessment value of a road event is greater than or equal to the intensity threshold, it is considered that there is a significant causal impact between nodes. Priority is given to tracing the source and locating key upstream nodes, dynamically optimizing traffic light timing, and coordinating emergency resources. Real-time risk warnings and detour suggestions are pushed to drivers through vehicle terminals and navigation platforms. The propagation link and impact range of the road event are displayed in real time, and the source and causal chain of the abnormal event are automatically marked. A source tracing report is generated to clarify the initiation point, path and impact range of the abnormality.

5. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process of predicting traffic conditions based on real-time and historical preprocessed multidimensional traffic data is as follows: Feature extraction and sample construction are performed on real-time and historical multidimensional traffic data and road event intensity values ​​as multidimensional spatiotemporal feature input datasets. The multidimensional spatiotemporal feature input datasets are trained using a long short-term memory network algorithm to construct a traffic state prediction model, which outputs prediction results for traffic flow, average vehicle speed, road events, and road event intensity values ​​at each traffic node in the future.

6. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process of comprehensively analyzing traffic control needs based on traffic condition prediction results is as follows: Obtain the predicted traffic flow of node i at future time from the output of the traffic state prediction model; obtain the road event intensity value of node i at future time as the predicted road event intensity value; obtain the predicted traffic flow of node i multiple times, and obtain the predicted variance of traffic flow of node i at future time by calculating the sample variance; The average vehicle speed of node i at future time points is obtained as the predicted average vehicle speed. The predicted queue fluctuation value is obtained by multiplying the sum of the predicted traffic flow, the predicted road event intensity value and constant 1, and the traffic flow prediction variance. The predicted control demand value is obtained by dividing the predicted queue fluctuation value by the sum of the predicted average vehicle speed value and constant 1. The system compares the predicted control demand value with the control threshold in real time. When the control demand value is less than the control threshold, the system automatically optimizes the signal timing according to the basic traffic flow and the prediction results. It continuously tracks the traffic status of nodes and continues to perform predicted control demand analysis. The system calculates the predicted control demand value three times in a row. If the predicted control demand value increases three times in a row, the system switches to a key intervention response. When the demand value is greater than or equal to the control threshold, the green light duration of the node is automatically extended, the signal phase is optimized, and the surrounding intersections are coordinated to carry out regional traffic management, giving priority to alleviating the high pressure section. Real-time risk alerts, detour and diversion suggestions are pushed through in-vehicle navigation and road guidance screens; Record the effects of intervention and output dynamic reports on high-pressure nodes and links.

7. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process for assessing traffic anomalies based on preprocessed multidimensional traffic data and traffic state prediction results is as follows: Obtain the traffic flow at node i at the current time t and the predicted traffic flow at future times. Calculate the difference between the traffic flow at node i at the current time t and the predicted traffic flow at future times, and take the absolute value to obtain the traffic flow change magnitude. Obtain the average vehicle speed at node i at the current time t and the predicted average vehicle speed at future times. Calculate the difference between the average vehicle speed at node i at the current time t and the predicted average vehicle speed at future times, and take the absolute value to obtain the average vehicle speed change magnitude. Obtain the road event intensity value at node i at the current time t. The queue mutation term is obtained by multiplying the traffic flow weight factor by the traffic flow change magnitude; the speed mutation term is obtained by multiplying the speed weight factor by the average speed change magnitude; and the event intensity term is obtained by multiplying the event intensity value of node i at the current time t by the event intensity weight factor. The queue mutation term, speed mutation term, and event intensity term are added together to obtain the anomaly comprehensive evaluation value.

8. The intelligent traffic management system based on spatiotemporal sequence features according to claim 1, characterized in that, The specific process of combining traffic anomaly assessment results with spatiotemporal causal chains to implement adaptive traffic control measures and obtain the final multidimensional traffic data after control is as follows: Based on the output of historical multidimensional traffic data and traffic condition prediction models, the hierarchical anomaly detection thresholds for each road segment, time period, and type are dynamically set and adjusted. The system compares the comprehensive anomaly assessment value with the anomaly detection threshold in real time. When the comprehensive anomaly assessment value is less than the anomaly detection threshold, it is considered that there is no anomaly at the current moment. Regular traffic monitoring and multi-dimensional traffic data collection are maintained, and traffic anomalies are continuously monitored. When the comprehensive anomaly assessment value is greater than or equal to the anomaly detection threshold, it is considered that there is an anomaly at the current moment. Combining the key causal nodes and chain reaction chains between road events, the system automatically backtracks along the traffic network to analyze whether the anomaly is local and independent, chain-like propagation, or large-scale spread, and assesses the scope of impact and affected nodes accordingly. The system automatically optimizes traffic light timing, extends the green time, pushes diversion, detour, and speed reduction suggestions to vehicles and navigation platforms, and coordinates emergency resources to respond in advance. It also updates the high-pressure node and link dynamic reports on the management platform simultaneously.

Images