Urban traffic flow pre-judgment system

Abstract

The invention discloses an urban traffic flow pre-judgment system, and the system comprises a multi-source sensing module which is disposed at a traffic node, and collects traffic flow time series data in real time; the multi-source sensing module is connected with the dynamic graph engine, and the dynamic graph engine comprises a time sequence correlation calculation unit which is used for generating a dynamic correlation weight between nodes based on a dynamic time warping algorithm; and a topology fusion unit. According to the urban traffic flow pre-judgment system provided by the embodiment of the invention, the urban traffic flow pre-judgment system effectively captures the time-space dependency relationship of the traffic flow through multi-source data fusion and dynamic graph modeling, and provides real-time data support for congestion early warning and signal control; and the decision support terminal assists a traffic management department to efficiently formulate a scheduling strategy through visualization and interpretability analysis, so that the urban traffic monitoring and regulation efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of traffic systems, and in particular to an urban traffic flow prediction system. Background Art

[0002] Current mainstream graph neural network models (such as GCN) for urban traffic flow prediction generally rely on fixed road network topology and treat road connectivity as static prior knowledge (e.g., constructing an adjacency matrix based on geographic distance or lane number). This severely limits their ability to capture dynamic temporal correlations in traffic flow. In urban traffic, the flow dependencies between nodes exhibit significant time-varying characteristics. For example, tidal commuting flows across regions during peak hours can create strong temporal similarities between distant nodes (e.g., traffic fluctuations between suburban residential areas and downtown business districts exhibit synchronized cycles). Traditional static graph structures are unable to quantify these dynamic correlations between non-adjacent nodes. Furthermore, during sudden congestion events, traffic transmission paths can violate the inherent constraints of the road network topology (e.g., sudden increases in traffic at non-adjacent nodes caused by detours). Static graph models, lacking a mechanism for calculating temporal similarity, struggle to update node association weights in real time. This results in the prediction model's inability to adapt to the dynamic evolution of traffic flow, leading to delayed congestion warnings and reduced prediction accuracy. Summary of the Invention

[0003] The present application aims to solve one of the technical problems in the related art at least to a certain extent.

[0004] To this end, one of the purposes of this application is to provide an urban traffic flow prediction system that effectively captures the spatiotemporal dependencies of traffic flows through multi-source data fusion and dynamic graph modeling, providing real-time data support for congestion warning and signal control; the decision support terminal assists traffic management departments in efficiently formulating scheduling strategies through visualization and explainable analysis, thereby improving the efficiency of urban traffic monitoring and regulation.

[0005] To achieve the above-mentioned objectives, the first embodiment of the present application proposes an urban traffic flow prediction system, including: a multi-source perception module, deployed at a traffic node, to collect traffic flow time series data in real time; a dynamic graph engine, the multi-source perception module is connected to the dynamic graph engine, and the dynamic graph engine includes: a time series correlation calculation unit, which generates dynamic correlation weights between nodes based on a dynamic time warping algorithm; a topology fusion unit, which fuses the static connection relationship of the road network with the dynamic correlation weights and outputs a spatiotemporal fusion graph; an intelligent prediction module, which loads a spatiotemporal graph neural network model, processes the spatiotemporal fusion graph to generate prediction results; and a communication relay module, which realizes data transmission between the multi-source perception module, the dynamic graph engine, and the intelligent prediction module.

[0006] According to the urban traffic flow prediction system of the embodiment of the present application, the urban traffic flow prediction system effectively captures the spatiotemporal dependencies of traffic flows through multi-source data fusion and dynamic graph modeling, providing real-time data support for congestion warning and signal control; the decision support terminal assists traffic management departments in efficiently formulating scheduling strategies through visualization and explainable analysis, thereby improving the efficiency of urban traffic monitoring and regulation.

[0007] In addition, the urban traffic flow prediction system proposed in this application may also have the following additional technical features:

[0008] In one embodiment of the present application, the multi-source perception module includes a geomagnetic sensor, a microwave radar, a video camera roll deployed at an intersection, and an OBU device deployed at a vehicle-mounted terminal. The geomagnetic sensor, the microwave radar, the video camera, and the OBU device collect data in real time and synchronize it to a communication relay module.

[0009] In one embodiment of the present application, the timing association calculation unit includes:

[0010] Periodic feature analysis subunit: decomposes the long-term trend and periodic fluctuation components of traffic flow time series, and uses the seasonal decomposition model to decompose the long-term trend component, periodic fluctuation component and random noise component of traffic flow time series;

[0011] Dynamic alignment subunit: matches non-equal-length timing patterns through a dynamic time warping algorithm, calculates the optimal time warping path for non-equal-length timing sequences between nodes through the dynamic time warping algorithm, and obtains the timing alignment distance;

[0012] Similarity conversion subunit: maps the temporal alignment distance to the [0,1] interval association weight.

[0013] In one embodiment of the present application, the spatiotemporal graph neural network model includes a fusion structure of at least three graph convolutional layers and gated recurrent units.

[0014] In one embodiment of the present application, the communication relay module supports the vehicle-road cooperative communication protocol.

[0015] In one embodiment of the present application, a streaming update control unit is further included:

[0016] Mutation detector: triggers an update when the node traffic change exceeds a preset threshold;

[0017] Local reconstructor: recalculates the dynamic association weights of the changed nodes;

[0018] Incremental Fusion: Updates the spatiotemporal fusion graph by combining the historical graph status.

[0019] In one embodiment of the present application, the preset threshold is adaptively set according to the standard deviation of historical flow fluctuations;

[0020] The preset threshold comprises the following steps:

[0021] S1 collects traffic data of the target node for the past 7 days and calculates the standard deviation σ of the time series;

[0022] The preset threshold of S2 is set to k·σ, where k is an adjustable coefficient with a default value of 2.5;

[0023] S3 updates the standard deviation σ and preset thresholds every 24 hours based on the latest data.

[0024] In one embodiment of the present application, the intelligent prediction module includes a latent association analysis unit for identifying at least one of the following patterns:

[0025] Inter-regional tidal commuting correlation is determined based on the correlation coefficient of traffic flows between nodes during the morning and evening peak hours being greater than 0.7;

[0026] Hotspot area radiation association, using the graph attention mechanism to identify the hub nodes that rank in the top 10% in terms of their impact on the traffic of surrounding nodes;

[0027] The fluctuation transmission correlation of transportation hubs is analyzed by using Granger causality test to analyze the predictive ability of hub node traffic changes on downstream nodes.

[0028] In one embodiment of the present application, a decision support terminal is further connected, including:

[0029] A dynamic graph semantic renderer, which uses WebGL technology to visualize the spatiotemporal fusion graph;

[0030] The association interpretability verifier analyzes the prediction results of the neural network model described in the spatiotemporal graph through SHAP values ​​and generates a node association impact report;

[0031] The prediction and warning signal generator generates a warning signal and pushes it to the traffic management platform when the predicted traffic volume exceeds 80% of the road capacity.

[0032] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0034] Figure 1 Schematic diagram of the overall urban traffic flow prediction system according to the present application;

[0035] Figure 2 This is a diagram of the spatiotemporal graph neural network architecture in the dynamic graph of the urban traffic flow prediction system according to the present application;

[0036] Figure 3 A diagram showing the working mechanism of the streaming update control unit in the urban traffic flow prediction system according to the present application;

[0037] Figure 4 This is a functional architecture diagram of a terminal in the urban traffic flow prediction system according to the present application;

[0038] Figure 5 A schematic diagram of module deployment in the urban traffic flow prediction system according to the present application;

[0039] Figure 6 for Figure 1 Middle split figure;

[0040] Figure 7 for Figure 1 Middle split diagram. DETAILED DESCRIPTION

[0041] The embodiments of the present application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application and are not to be construed as limiting the present application. On the contrary, the embodiments of the present application include all variations, modifications, and equivalents that fall within the spirit and scope of the appended claims.

[0042] The following is combined with Figure 1 -Attached Figure 7 The urban traffic flow prediction system of the embodiment of the present application is described as follows, including:

[0043] Multi-source perception modules are deployed at traffic nodes to collect traffic flow time series data in real time;

[0044] The dynamic graph engine, the multi-source perception module is connected to the dynamic graph engine, and the dynamic graph engine includes:

[0045] The time series correlation calculation unit generates dynamic correlation weights between nodes based on the dynamic time warping algorithm;

[0046] The topology fusion unit integrates the static connectivity of the road network with the dynamic association weights and outputs a spatiotemporal fusion graph. The static connectivity of the road network is represented by a directed graph, where the node set corresponds to the traffic node, the edge set corresponds to the road connection, and the edge attributes include static parameters such as the number of lanes, speed limit, and road length.

[0047] The spatiotemporal fusion graph is a dynamic graph with weights. The node status contains real-time traffic flow characteristics, and the edge weight is generated by fusing the static connection relationship weight and the dynamic association weight according to the linear weighted fusion model, where α is an adjustable weight coefficient of 0-1.

[0048] The intelligent prediction module loads the spatiotemporal graph neural network model and processes the spatiotemporal fusion graph to generate prediction results;

[0049] The communication relay module realizes data transmission between the multi-source perception module, dynamic graph engine, and intelligent prediction module.

[0050] The dynamic graph engine's time series correlation calculation unit uses the DTW algorithm to solve the problem of unequal time series lengths at different nodes caused by differences in sampling frequency or data missing. By dynamically aligning time series and calculating distances, it converts the temporal correlation of traffic flow into a quantifiable weight value. For example, the more similar the traffic fluctuation patterns at two intersections during peak hours in the morning and evening, the higher the dynamic correlation weight, providing a dynamic correlation basis for spatiotemporal fusion.

[0051] The static connection relationship of the road network in the topological fusion unit is modeled using a directed graph to accurately depict the road's traffic direction, such as one-way streets, turning restrictions and physical properties. For example, the number of lanes affects traffic capacity and speed limits affect traffic speed. The dynamic association weight makes up for the real-time traffic interaction that the static road network model cannot capture, such as the sudden increase in cross-regional traffic caused by emergencies. The two achieve adaptability to different traffic scenarios through the adjustable weight coefficient α. For example, the static weight ratio can be increased in daily commuting scenarios, and the impact of dynamic association can be enhanced in sudden congestion scenarios.

[0052] The generated spatiotemporal fusion graph serves as the input of the intelligent prediction module. Its node status integrates multi-dimensional traffic flow characteristics collected in real time, such as current traffic volume and average vehicle speed. The edge weight also includes the inherent connection strength of the road network and the real-time dynamic correlation, providing a feature representation that is both structural and timely for the spatiotemporal graph neural network, supporting the model to efficiently learn the spatiotemporal evolution laws of traffic flow.

[0053] The deployment of multi-source perception modules covers key traffic nodes in the city, and can realize multi-dimensional data collection through geomagnetic sensors, microwave radars, video cameras and other devices to ensure the integrity of input data; the communication relay module must support low-latency data transmission protocols such as 5G and C-V2X to meet the stringent real-time requirements of traffic flow prediction.

[0054] In one embodiment of the present application, the multi-source perception module includes a geomagnetic sensor, a microwave radar, a video camera roll deployed at an intersection, and an OBU device deployed at a vehicle-mounted terminal. The geomagnetic sensor, microwave radar, video camera, and OBU device collect data in real time and synchronize it to the communication relay module.

[0055] Deployed at urban traffic nodes such as intersections, bridges, and expressway ramps, it uses heterogeneous sensors to collect multi-dimensional traffic flow time series data in real time, including:

[0056] Basic traffic data: node traffic volume, lane occupancy, queue length;

[0057] Dynamic speed data: average speed, speed variance;

[0058] Spatiotemporal correlation data: the ratio of turning traffic between adjacent nodes, such as the proportion of straight vehicles at intersection A entering intersection B.

[0059] It supports geomagnetic sensors, vehicle passing detection, microwave radar, non-contact vehicle speed monitoring, video cameras, visual recognition of vehicle types and traffic flow, and other equipment. The sampling frequency is not less than 500ms to ensure that the data granularity meets short-term prediction needs.

[0060] Sensor configuration:

[0061] Geomagnetic sensor: deployed 5m before the stop line, detecting the time interval between vehicles passing through and calculating traffic flow;

[0062] Microwave radar: installed on light poles, covering an area with a radius of 50m at the intersection, and monitoring the speed of vehicles in each lane in real time;

[0063] Video camera: A 2-megapixel smart dome camera uses a deep learning algorithm to identify vehicle type classification and queue length.

[0064] In one embodiment of the present application, the timing correlation calculation unit includes:

[0065] Periodic feature analysis subunit: decomposes the long-term trend and periodic fluctuation components of traffic flow time series, and uses the seasonal decomposition model STL or Fourier transform to decompose the long-term trend component, periodic fluctuation component and random noise component of traffic flow time series;

[0066] Dynamic alignment subunit: matches non-uniform time series patterns using the dynamic time warping algorithm, calculates the optimal time warping path of non-uniform time series between nodes using the dynamic time warping algorithm DTW, and obtains the time series alignment distance;

[0067] Similarity conversion subunit: maps the time series alignment distance to the association weight in the [0,1] interval, and normalizes the time series alignment distance to the association weight in the [0,1] interval through the formula, where is the historical maximum value of the node pair time series distance.

[0068] Specifically, the periodic feature analysis subunit, wherein,

[0069] 1) Long-term trend component: captures the changing patterns of traffic flow over a longer time scale, such as the continuous growth of traffic flow on weekdays and the overall decrease of traffic flow on holidays;

[0070] 2) Periodic fluctuation component: extract regular fluctuations with fixed periods, such as the 24-hour period of morning and evening peaks, and the 7-day period during the week and weekends;

[0071] 3) Random noise component: Filter out sudden outliers or measurement errors, such as instantaneous traffic changes caused by temporary traffic accidents.

[0072] By decomposing and removing low-frequency trends and noise in the data, high-frequency change features that reflect real-time dynamic interactions are retained, providing purer input for subsequent time series alignment.

[0073] Dynamic alignment subunit: For time series with unequal lengths due to sampling frequency differences, data missing, etc. at different nodes, the dynamic time warping algorithm DTW is used to calculate the optimal time warping path:

[0074] 1) In the calculation of time series similarity, local time axis stretching or compression is allowed, such as aligning the 1-minute interval A node sequence with the 2-minute interval B node sequence;

[0075] 2) By minimizing the cumulative distance criterion, the best matching path between the two sequences is found, and the time sequence alignment distance dij is output. The smaller the distance, the more similar the traffic flow time sequence patterns of the two nodes are, such as the high synchronization of traffic fluctuations at adjacent intersections.

[0076] Similarity conversion subunit: normalizes the temporal alignment distance dij calculated by DTW to a dynamic association weight formula in the interval [0,1].

[0077] 1) In the dynamic association weight formula, max(dij) is the maximum value of the historical time series distance between nodes;

[0078] 2) Normalization makes the association strength of different node pairs comparable. The closer the weight value is to 1, the stronger the dynamic association between nodes. For example, the weight of nodes on the commuting route during the morning rush hour is close to 1.

[0079] In one embodiment of the present application, the spatiotemporal graph neural network model includes a fusion structure of at least three layers of graph convolutional layers GCN and gated recurrent units GRU.

[0080] The input layer receives the node feature matrix and adjacency matrix of the spatiotemporal fusion graph, extracts hierarchical features, and outputs the multi-period traffic flow prediction values ​​for the next 15 minutes to 1 hour through the fully connected layer.

[0081] Specifically, the three-layer GCN structure aggregates neighbor node information layer by layer, such as the traffic flow at intersection A is affected by adjacent intersections B and C, and extracts road network spatial dependency features, such as congestion propagation paths.

[0082] GRU time series modeling: Captures the time series dynamics of each node, such as the periodic changes in morning and evening peaks, and selectively retains historical information through a gating mechanism, updating gates, and resetting gates.

[0083] Multi-scale prediction output: After integrating spatiotemporal features through the fully connected layer, it outputs prediction values ​​for multiple time steps such as 15 minutes, 30 minutes, and 1 hour in the future, covering both short-term traffic situation awareness and medium- and long-term decision support needs.

[0084] In one embodiment of the present application, the communication relay module supports the vehicle-road cooperative communication protocol.

[0085] The communication relay module supports DSRC, dedicated short-range communication, C-V2X, cellular vehicle-road collaboration and 5G-NR communication protocols, and uses MQTT message queues to achieve data access and distribution of heterogeneous devices, with data transmission delay controlled within 200ms.

[0086] In one embodiment of the present application, a streaming update control unit is further included:

[0087] Mutation detector: triggers an update when the node traffic change exceeds a preset threshold. It captures abnormal events such as traffic accidents in real time by setting a preset threshold, such as when the traffic change exceeds 2.5 times the historical standard deviation.

[0088] Local reconstructor: recalculates the dynamic association weights of changed nodes to avoid global computation overhead;

[0089] Incremental Fusion: Updates the spatiotemporal fusion graph based on the historical graph status to maintain the model's ability to respond quickly to real-time changes.

[0090] In one embodiment of the present application, the preset threshold is adaptively set according to the standard deviation of historical traffic fluctuations.

[0091] The preset threshold is adaptively set through the following steps:

[0092] 1) Collect traffic data of the target node for the past 7 days, calculate the standard deviation σ of the time series, and quantify the historical fluctuation range;

[0093] 2) The preset threshold is set to k·σ, where k is an adjustable coefficient with a default value of 2.5, k times the standard deviation, and the default value of k=2.5 is used as the critical value of traffic change that triggers the update, balancing the sensitivity and stability of anomaly detection;

[0094] 3) Update the standard deviation σ and preset thresholds based on the latest data every 24 hours to adapt to periodic fluctuations in traffic flow, such as differences between weekdays and weekends.

[0095] In one embodiment of the present application, the intelligent prediction module includes a latent association analysis unit for identifying at least one of the following patterns:

[0096] Inter-regional tidal commuting correlation is determined based on the correlation coefficient of traffic flows between nodes during the morning and evening peak hours being greater than 0.7;

[0097] Hotspot area radiation association, using the graph attention mechanism GAT to identify the top 10% hub nodes that influence the traffic of surrounding nodes;

[0098] The transmission correlation of transportation hub fluctuations was analyzed by using Granger causality test to analyze the predictive ability of hub node traffic changes on downstream nodes, with a significance level better than 0.05.

[0099] In one embodiment of the present application, a decision support terminal is further connected, including:

[0100] Dynamic graph semantic renderer, using WebGL technology to visualize spatiotemporal fusion graphs;

[0101] The association interpretability verifier analyzes the prediction results of the spatiotemporal graph neural network model through SHAP values ​​and generates a node association impact report;

[0102] The prediction and warning signal generator generates warning signals according to the three-level standards of red > 90%, yellow 80%-90%, and blue < 80% and pushes them to the traffic management platform when the predicted traffic volume exceeds 80% of the road capacity.

[0103] In summary, the urban traffic flow prediction system of the embodiment of the present application, in the actual operation process of the urban traffic flow prediction system,

[0104] 1. Data collection and access.

[0105] The multi-source perception module uses geomagnetic sensors, microwave radars, video cameras and on-board OBU equipment to collect traffic flow data, including flow rate, vehicle speed, occupancy rate, etc. in real time at a sampling frequency of ≥500ms, and synchronizes it to the communication relay module.

[0106] The communication relay module transmits heterogeneous device data to the dynamic graphics engine with low latency through DSRC / C-V2X / 5G-NR protocols and MQTT queues.

[0107] 2. Dynamic graph modeling and updating.

[0108] Timing correlation calculation unit:

[0109] Use STL / Fourier transform to decompose the trend, cycle and noise components of time series data;

[0110] Based on the DTW algorithm, it aligns time series of different lengths and calculates the dynamic association weights between nodes, such as the similarity of morning and evening peak traffic.

[0111] Topological fusion unit:

[0112] The static directed graph of the road network, the number of lanes, speed limits, etc., are integrated with dynamic weights to generate a spatiotemporal fusion graph through a linear model.

[0113] Streaming update control: The mutation detector monitors traffic mutations based on an adaptive threshold of historical standard deviation;

[0114] After triggering, the local reconstructor recalculates the association weights of the changed nodes, and the incremental fusion updates the space-time graph.

[0115] 3. Intelligent prediction and analysis.

[0116] Spatiotemporal graph neural network: input the node feature matrix and adjacency matrix of the spatiotemporal fusion graph;

[0117] The 3-layer GCN extracts spatial dependency congestion propagation, and the GRU captures temporal cycles;

[0118] The fully connected layer outputs the predicted values ​​of traffic flow, vehicle speed, etc. for multiple time periods in the future 15 minutes to 1 hour.

[0119] Hidden association analysis.

[0120] 1) Tidal commuting: A correlation coefficient > 0.7 determines regional linkage;

[0121] 2) Hotspot Radiation: GAT screens the top 10% of hub nodes;

[0122] 3). Hub transmission: Granger causality test p<0.05 analyzes fluctuation transmission.

[0123] 4. Early warning.

[0124] The dynamic graph semantic renderer uses WebGL to visualize spatiotemporal graphs, showing traffic heatmaps and correlation strength;

[0125] The associated interpretability verifier generates node influence reports using SHAP values;

[0126] When the predicted traffic volume exceeds 80% of the road capacity, a warning signal will be generated according to the three-level standards of red > 90%, yellow 80%-90%, and blue < 80% and pushed to the traffic management platform.

[0127] The urban traffic flow prediction system of this application effectively captures the spatiotemporal dependencies of traffic flows through multi-source data fusion and dynamic graph modeling, providing real-time data support for congestion warning and signal control; the decision support terminal assists traffic management departments in efficiently formulating scheduling strategies through visualization and explainable analysis, thereby improving the efficiency of urban traffic monitoring and regulation.

[0128] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0129] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0130] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and deform the above embodiments within the scope of the present application.

Claims

1. An urban traffic flow prediction system, characterized in that: include: Multi-source perception modules are deployed at traffic nodes to collect traffic flow time series data in real time; A dynamic graph engine, wherein the multi-source perception module is connected to the dynamic graph engine, and the dynamic graph engine includes: The time series correlation calculation unit generates dynamic correlation weights between nodes based on the dynamic time warping algorithm; A topology fusion unit, which fuses the static connection relationship of the road network with the dynamic association weight and outputs a spatiotemporal fusion graph; An intelligent prediction module, which loads the spatiotemporal graph neural network model and processes the spatiotemporal fusion graph to generate prediction results; The communication relay module realizes data transmission between the multi-source perception module, the dynamic graph engine, and the intelligent prediction module.

2. The urban traffic flow prediction system according to claim 1, characterized in that: The multi-source perception module includes a geomagnetic sensor, a microwave radar, a video camera roll deployed at the intersection and an OBU device deployed at the vehicle terminal. The geomagnetic sensor, the microwave radar, the video camera and the OBU device collect data in real time and synchronize them to the communication relay module.

3. The urban traffic flow prediction system according to claim 1, characterized in that: The time series correlation calculation unit includes: Periodic feature analysis subunit: decomposes the long-term trend and periodic fluctuation components of traffic flow time series, and uses the seasonal decomposition model to decompose the long-term trend component, periodic fluctuation component and random noise component of traffic flow time series; Dynamic alignment subunit: matches non-equal-length timing patterns through a dynamic time warping algorithm, calculates the optimal time warping path for non-equal-length timing sequences between nodes through the dynamic time warping algorithm, and obtains the timing alignment distance; Similarity conversion subunit: maps the temporal alignment distance to the [0,1] interval association weight.

4. The urban traffic flow prediction system according to claim 1, characterized in that: The spatiotemporal graph neural network model includes a fusion structure of at least three graph convolutional layers and gated recurrent units.

5. The urban traffic flow prediction system according to claim 1, characterized in that: The communication relay module supports the vehicle-road cooperative communication protocol.

6. The urban traffic flow prediction system according to claim 1, characterized in that: Also includes streaming update control unit: Mutation detector: triggers an update when the node traffic change exceeds a preset threshold; Local reconstructor: recalculates the dynamic association weights of the changed nodes; Incremental Fusion: Updates the spatiotemporal fusion graph by combining the historical graph status.

7. The urban traffic flow prediction system according to claim 6, characterized in that: The preset threshold is adaptively set according to the standard deviation of historical traffic fluctuations; The preset threshold comprises the following steps: S1 collects traffic data of the target node for the past 7 days and calculates the standard deviation σ of the time series; The preset threshold of S2 is set to k·σ, where k is an adjustable coefficient with a default value of 2.5; S3 updates the standard deviation σ and preset thresholds every 24 hours based on the latest data.

8. The urban traffic flow prediction system according to claim 1, characterized in that: The intelligent prediction module includes a hidden association analysis unit for identifying at least one of the following patterns: Inter-regional tidal commuting correlation is determined based on the correlation coefficient of traffic flows between nodes during the morning and evening peak hours being greater than 0.7; Hotspot area radiation association, using the graph attention mechanism to identify the hub nodes that rank in the top 10% in terms of their impact on the traffic of surrounding nodes; The fluctuation transmission correlation of transportation hubs is analyzed by using Granger causality test to analyze the predictive ability of hub node traffic changes on downstream nodes.

9. The urban traffic flow prediction system according to claim 1, characterized in that: Also connected to decision support terminals, including: A dynamic graph semantic renderer, which uses WebGL technology to visualize the spatiotemporal fusion graph; The association interpretability verifier analyzes the prediction results of the neural network model described in the spatiotemporal graph through SHAP values ​​and generates a node association impact report; The prediction and warning signal generator generates a warning signal and pushes it to the traffic management platform when the predicted traffic volume exceeds 80% of the road capacity.

Images