A traffic control management method and system based on distributed architecture multi-source big data

Abstract

The embodiment of the present application discloses a traffic control management method and system based on distributed architecture multi-source big data, which belongs to the field of smart transportation technology. It solves the problem that the existing technology has insufficient accuracy in predicting traffic conditions, which easily leads to inaccurate traffic control. It includes: constructing a dynamic traffic flow model based on the extracted feature data and the dynamic spatiotemporal graph corresponding to the traffic network; deducing the traffic situation through the central node in the traffic network to obtain a variety of traffic prediction scenarios, and matching the corresponding traffic control strategies; after detecting an abnormal traffic event, delineating the affected area according to the corresponding propagation impact trend of the abnormal traffic event on the spatial axis; dynamically combining multiple operating units corresponding to the traffic control strategy through the central node to generate a customized control plan; each edge node jointly optimizes the customized control plan based on the local traffic flow data using a distributed optimization algorithm to achieve traffic control management.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a method and system for traffic control management based on distributed architecture and multi-source big data. Background Art

[0002] With the acceleration of urbanization and the continuous expansion of transportation networks, traditional traffic control and management methods are gradually becoming limited in dealing with complex traffic scenarios. Currently, most traffic control systems rely on single-point data collection and centralized processing. This centralized architecture suffers from high computational load and high response latency when processing multi-source, heterogeneous traffic data. Especially in large-scale road networks, data transmission pressure and processing bottlenecks at central nodes are significant, making it difficult to meet the needs of real-time control.

[0003] Secondly, although some solutions in the existing technology introduce historical data modeling, it is difficult to capture the dynamically changing traffic flow characteristics in the road network in real time. Therefore, the existing technology still uses preset fixed control strategies to control the road network.

[0004] However, during peak hours in the morning and evening or sudden abnormal events, fixed control strategies are difficult to fully grasp the spatiotemporal propagation laws of traffic conditions, resulting in insufficient prediction accuracy, which can easily lead to low traffic control accuracy. Summary of the Invention

[0005] The embodiments of the present application provide a multi-source big data traffic control management method and system based on a distributed architecture, which is used to solve the following technical problems: during rush hours in the morning and evening or sudden abnormal events, fixed control strategies are difficult to fully grasp the spatiotemporal propagation laws of traffic conditions, resulting in insufficient prediction accuracy, which can easily lead to inaccurate traffic control.

[0006] The embodiments of this application adopt the following technical solutions:

[0007] The embodiment of the present application provides a traffic control management method based on a distributed architecture and multi-source big data. The method comprises: extracting feature data from historical traffic flow data corresponding to a traffic network based on a time sequence, and constructing a dynamic traffic flow model based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network; using a central node in the traffic network, and based on the dynamic traffic flow model, predicting the traffic situation within a preset time period in the future, obtaining multiple traffic prediction scenarios, and matching corresponding traffic control strategies for the multiple traffic prediction scenarios; after detecting an abnormal traffic event, demarcating the affected area based on the propagation impact trend of the abnormal traffic event on the spatial axis with the road section corresponding to the abnormal traffic event as the center; dividing the traffic control strategy into multiple operation units through the central node, and dynamically combining the operation units based on the traffic information corresponding to the abnormal traffic event to generate a customized control plan; determining the edge nodes corresponding to the affected area, and determining the local traffic flow data corresponding to the edge nodes, and each edge node jointly optimizing the customized control plan based on the local traffic flow data using a distributed optimization algorithm to achieve traffic control management.

[0008] In one implementation of the present application, feature data is extracted from historical traffic flow data corresponding to a traffic network based on a time sequence, and a dynamic traffic flow model is constructed based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network. Specifically, the method includes: merging the historical traffic flow data based on the timestamps corresponding to the historical traffic flow data and the location information of the traffic network; extracting key data from the merged historical traffic flow data through a sliding time window, and determining the dynamic traffic change characteristics corresponding to the historical traffic flow data through a preset temporal convolutional network based on the extracted key data; determining the spatiotemporal correlation entropy corresponding to the historical traffic flow data, comparing the spatiotemporal correlation entropy with the preset spatiotemporal correlation entropy, and extracting abnormal traffic characteristics based on the comparison results; constructing a directed weighted graph based on the traffic network, and dynamically updating the traffic flow state attributes of the directed weighted graph to form a dynamic traffic spatiotemporal graph sequence; and constructing a dynamic traffic flow model based on the dynamic traffic change characteristics, abnormal traffic characteristics, and the dynamic traffic spatiotemporal graph sequence.

[0009] In one implementation of the present application, a dynamic traffic flow model is constructed based on dynamic traffic change characteristics, abnormal traffic characteristics and a dynamic traffic spatiotemporal graph sequence, specifically including: constructing a multidimensional feature matrix based on the dynamic traffic change characteristics, abnormal traffic characteristics and a dynamic traffic spatiotemporal graph sequence; determining the hidden spatial associations between road network nodes corresponding to the dynamic traffic spatiotemporal graph sequence based on the multidimensional feature matrix, and extracting multi-scale spatial features by stacking multiple layers of graph convolution; processing the multidimensional feature matrix through a temporal convolutional network and traffic spatiotemporal coding to determine the time-dependent features corresponding to the dynamic traffic change characteristics; wherein the traffic spatiotemporal coding is a time position identifier generated for different traffic time periods by sine-cosine functions; generating gating weights corresponding to abnormal traffic characteristics based on spatiotemporal correlation entropy, and weightedly integrating the abnormal traffic characteristics, multi-scale spatial features and time-dependent features through the gating weights; inputting the integrated feature data into a multi-head self-attention mechanism to determine the cross-spatiotemporal correlation weights between different road sections and / or different time periods to construct a dynamic traffic flow model.

[0010] In one implementation of the present application, a central node in a traffic network is used to deduce and predict traffic conditions within a preset future time period based on a dynamic traffic flow model, and a variety of traffic prediction scenarios are obtained, specifically including: constructing a federated learning network corresponding to traffic flow data based on the central node, edge nodes, and a dynamic traffic flow model, and embedding a confidence interval prediction network in the dynamic traffic flow model; wherein the federated learning network is a hierarchical distributed architecture, and the confidence interval prediction network is used to output the confidence interval of the prediction result; receiving traffic flow data sent by each edge node in the federated learning network through the central node, and updating the weight probability distribution of the confidence interval prediction network through the federated averaging algorithm; inputting traffic flow data into the dynamic traffic flow model to deduce and predict traffic conditions within a preset future time period, and generating a variety of traffic prediction scenarios with confidence intervals through Monte Carlo sampling; propagating the prediction confidence interval of each area along the road section to the edge node through the central node, and triggering a high-risk scenario identification when the width of the prediction confidence interval corresponding to any area does not meet the preset threshold.

[0011] In one implementation of the present application, corresponding traffic control strategies are matched for a variety of traffic prediction scenarios, specifically including: dividing data of similar traffic scenarios in historical traffic data, and establishing a corresponding control strategy library for each scenario category; obtaining a plurality of traffic prediction scenarios within a preset time period in the future through a dynamic traffic flow model, calculating similarity between the plurality of traffic prediction scenarios and the historical scenario categories, and selecting matching scenario sets corresponding to each traffic prediction scenario based on the similarity; constructing a first timing curve based on the historical traffic flow data corresponding to each matching scenario set, and constructing a second timing curve based on the traffic data flow corresponding to the traffic prediction scenario; selecting a candidate scenario from the matching scenario set based on the amplitude difference between the first timing curve and the second timing curve; extracting a candidate strategy set from the control strategy library corresponding to each candidate scenario, and screening the traffic control strategies corresponding to each traffic prediction scenario in the candidate strategy set with the road network delay rate as the target.

[0012] In one implementation of the present application, after an abnormal traffic event is detected, the affected area is delineated based on the propagation impact trend of the abnormal traffic event on a spatial axis, with the road section corresponding to the abnormal traffic event as the center. Specifically, the following steps are performed: after the abnormal traffic event is detected, traffic information corresponding to the abnormal traffic event is obtained; wherein the traffic information includes at least the type of traffic event, the time of occurrence of the event, and the traffic volume corresponding to the road section; the spatial axis is constructed with the road section corresponding to the abnormal traffic event as the center and the road sections directly or indirectly connected to the road section as adjacent nodes;

[0013] Traffic information and spatial axes are input into a preset neural network to output the propagation probability and propagation impact attenuation value of abnormal traffic events to adjacent nodes; based on the comparison results between the propagation probability and the preset probability threshold, a reference spatial area set is determined in the spatial axis; and based on the surrounding land use type corresponding to each road section and the historical event response frequency, the traffic sensitivity of each road section at the time of the traffic event is determined; based on the propagation impact attenuation value and traffic sensitivity, the reference spatial area set is screened to obtain the affected area.

[0014] In one implementation of the present application, a traffic control strategy is divided into multiple operation units through a central node, and the operation units are dynamically combined based on the traffic information corresponding to the abnormal traffic event, specifically including: extracting a traffic plan that matches the abnormal traffic event from the pre-generated traffic control strategy through the central node; dividing the traffic plan into multiple operation units through the central node, annotating the execution conditions, dependencies and conflict rules for each operation unit, and entering the traffic knowledge graph; performing semantic matching and spatiotemporal constraint filtering on the traffic information corresponding to the abnormal traffic event and the traffic knowledge graph through the central node to screen out a candidate set of applicable operation units; dynamically combining the operation units in the candidate set of operation units based on a preset objective function and optimization strategy to generate a customized control plan corresponding to the abnormal traffic event; the central node performs collaborative relationship verification and conflict detection on the customized control plan through the relationship network corresponding to the traffic knowledge graph to dynamically correct the customized control plan.

[0015] In one implementation of the present application, through a central node, based on a preset objective function and an optimization strategy, the operation units in the operation unit candidate set are dynamically combined to generate a customized control plan corresponding to the abnormal traffic event, specifically including: constructing a multidimensional objective function of the traffic control plan through the central node; wherein the multidimensional objective function includes efficiency objectives and resource objectives; through the central node, the operation unit candidate set is screened at the first level based on the constraint relationship of the traffic knowledge graph, and the operation unit candidate set is screened at the second level according to the urgency score; the screened operation unit candidate set is iteratively selected through the greedy algorithm and the multidimensional objective function until the preset termination condition is met to obtain an initial combination plan; wherein the preset termination condition includes the objective function improvement value being less than a threshold or resource exhaustion; the initial combination plan generated by the greedy algorithm is optimized through the reinforcement learning model in the central node; the greedy algorithm and the reinforcement learning model are dynamically weighted according to the event development stage; and the greedy algorithm and the reinforcement learning model are reintegrated according to the adjusted weights to obtain customized control plans corresponding to different event development stages.

[0016] In one implementation of the present application, each edge node jointly optimizes the customized control scheme based on local traffic flow data using a distributed optimization algorithm, specifically including: the central node collects the global road network status, determines the influence weight of each road section, and sends the influence weight and the global optimization target to the corresponding edge node; the edge node iteratively optimizes the operating unit parameters in the customized control scheme through a distributed optimization algorithm based on constraints, local traffic flow data and global optimization targets; adjacent edge nodes exchange the optimization results of boundary intersections and eliminate parameter conflicts through a negotiation mechanism; the edge node reports the optimized control scheme to the central node, and the central node verifies the global consistency and then sends it for execution.

[0017] The present application provides a multi-source big data traffic control and management system based on a distributed architecture, characterized in that the system includes: an extraction unit that extracts feature data from historical traffic flow data corresponding to a traffic network based on a time sequence, and constructs a dynamic traffic flow model based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network; a prediction unit that, through a central node in the traffic network, deduces and predicts traffic conditions within a preset future time period based on the dynamic traffic flow model, obtains multiple traffic prediction scenarios, and matches corresponding traffic control strategies for the multiple traffic prediction scenarios; a division unit that, after detecting an abnormal traffic event, demarcates an affected area based on the propagation impact trend of the abnormal traffic event on a spatial axis, with the road section corresponding to the abnormal traffic event as the center; a combination unit that divides the traffic control strategy into multiple operation units through the central node, and dynamically combines the operation units based on traffic information corresponding to the abnormal traffic event to generate a customized control plan; and an optimization unit that determines edge nodes corresponding to the affected area and local traffic flow data corresponding to the edge nodes, and each edge node jointly optimizes the customized control plan based on the local traffic flow data using a distributed optimization algorithm to achieve traffic control management.

[0018] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: The embodiments of the present application construct a model based on the temporal characteristics of historical traffic flow data and the dynamic spatiotemporal graph of the road network, which can capture the periodic fluctuations of traffic flow in the time dimension and the congestion propagation correlation in the spatial dimension, and more accurately deduce future traffic situations than traditional static models. By generating a variety of traffic prediction scenarios and pre-matching differentiated control strategies for each scenario, a linkage response between prediction and strategy is achieved, avoiding the lag of traditional fixed strategies. The affected areas are delineated according to the spatial propagation trend, and the upstream and downstream sections that need to be linked and regulated are accurately determined to reduce the waste of control resources. The control strategy is split into operation units and dynamically combined, and distributed joint optimization is performed in combination with the local traffic flow data of the edge nodes, which not only reduces the computing pressure of the central node, but also can quickly adjust the strategy according to the local real-time conditions, thereby improving the response speed and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present application. For those skilled in the art, other drawings can be obtained based on these drawings without inventive work. In the drawings:

[0020] Figure 1 A flow chart of a traffic control management method based on distributed architecture multi-source big data provided in an embodiment of the present application;

[0021] Figure 2 A schematic diagram of a multi-source big data traffic control management system based on a distributed architecture provided in an embodiment of the present application.

[0022] Reference numerals:

[0023] 200: Multi-source big data traffic control and management system based on distributed architecture, 201: extraction unit, 202: prediction unit, 203: division unit, 204: combination unit, 205: optimization unit. DETAILED DESCRIPTION

[0024] The embodiments of the present application provide a method and system for traffic control management based on a distributed architecture and multi-source big data.

[0025] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0026] The technical solutions proposed in the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] Figure 1 A flow chart of a multi-source big data traffic control management method based on a distributed architecture is provided in the embodiment of the present application. Figure 1 As shown, the traffic control management method based on distributed architecture multi-source big data includes the following steps:

[0028] Step 101: Extract feature data from historical traffic flow data corresponding to the traffic network based on a time sequence, and construct a dynamic traffic flow model based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network.

[0029] In one implementation of the present application, historical traffic flow data is merged based on the timestamps corresponding to the historical traffic flow data and the location information of the traffic network. Key data of the merged historical traffic flow data is extracted through a sliding time window, and based on the extracted key data, the dynamic traffic change characteristics corresponding to the historical traffic flow data are determined through a preset time series convolutional network. The spatiotemporal correlation entropy corresponding to the historical traffic flow data is determined, the spatiotemporal correlation entropy is compared with the preset spatiotemporal correlation entropy, and abnormal traffic characteristics are extracted based on the comparison results. A directed weighted graph is constructed based on the traffic network, and the traffic flow state attributes of the directed weighted graph are dynamically updated to form a dynamic traffic spatiotemporal graph sequence. A dynamic traffic flow model is constructed based on the dynamic traffic change characteristics, abnormal traffic characteristics, and the dynamic traffic spatiotemporal graph sequence.

[0030] Specifically, based on the timestamps in historical traffic flow data and location information on the road network, such as road segment IDs and coordinates, the flow and speed data for adjacent road segments within the same time period are merged into a single record, ensuring both temporal continuity and spatial correlation. A sliding time window mechanism is used to process the merged historical data. As the window slides along the time axis, it automatically filters out key data within each time interval, such as peak flow rate, average speed, and congestion index, to capture the dynamic changes in traffic flow over short periods. This extracted key data is input into a pre-configured Temporal Convolutional Network (TCN). This network, through a causal convolutional structure and dilated convolutional layers, automatically learns the dependencies of traffic flow over time series. The training process for this TCN is as follows: the input samples are fixed-length time series segments extracted from historical traffic data, covering key data such as peak flow rate, average speed, and congestion index; the output samples are the actual traffic flow status data for the corresponding time series, such as the actual flow rate or congestion status at a future moment. The input sample is passed into the time series convolutional network for forward propagation to obtain the predicted value, compared with the actual value to calculate the loss function, and then the network parameters are updated through back propagation, and the training is completed through repeated iterative optimization.

[0031] Furthermore, the spatiotemporal correlation entropy of historical traffic flow data is calculated, quantifying the uncertainty and correlation of traffic flow in both temporal and spatial dimensions using information entropy theory. Specifically, the temporal entropy is first calculated. Based on the probability distribution of traffic flow data for each time period, such as flow rate and speed, the entropy value is calculated to reflect the uncertainty of the time series. The spatial entropy is then calculated. The joint entropy is calculated based on the correlation matrix of data between road segments, reflecting the degree of spatial correlation. Finally, the spatiotemporal correlation entropy is calculated using the joint probability distribution of the spatiotemporal dimensions. The higher the value, the stronger the spatiotemporal uncertainty and correlation of traffic flow. For example, under normal circumstances, traffic flow changes between adjacent road segments are highly correlated, resulting in a low entropy value. However, sudden accidents can cause a sudden increase in the correlation entropy. The calculated spatiotemporal correlation entropy is compared with a preset normal entropy range. If it exceeds a threshold, it is identified as an anomaly, and the anomaly features are then extracted. A directed weighted graph is constructed based on the traffic network: nodes represent road segments, edges represent traffic flow directions between road segments, and weights are determined by attributes such as historical flow rate and travel time. The weights of the edges in the graph are dynamically adjusted according to the real-time updated traffic flow data. For example, the weight of travel time is increased during congestion, thus forming a dynamic traffic spatiotemporal graph sequence that changes over time.

[0032] Furthermore, the dynamic traffic change characteristics extracted by the time series convolutional network and the abnormal traffic characteristics obtained by spatiotemporal correlation entropy analysis are integrated with the dynamically updated traffic spatiotemporal graph sequence to construct a dynamic traffic flow model. This model can simultaneously capture the normal dynamic patterns of traffic flow, abnormal and sudden characteristics, and the spatiotemporal correlation of road network topology, achieving accurate modeling and prediction of traffic trends.

[0033] In one implementation of the present application, a multi-dimensional feature matrix is ​​constructed based on dynamic traffic change characteristics, abnormal traffic characteristics and dynamic traffic spatiotemporal graph sequences. Based on the multi-dimensional feature matrix, the hidden spatial associations between the road network nodes corresponding to the dynamic traffic spatiotemporal graph sequence are determined, and multi-scale spatial features are extracted by stacking multiple layers of graph convolution. The multi-dimensional feature matrix is ​​processed by a temporal convolutional network and traffic spatiotemporal coding to determine the time-dependent features corresponding to the dynamic traffic change characteristics; wherein, the traffic spatiotemporal coding is a time position identifier generated for different traffic time periods by sine-cosine functions. Gating weights corresponding to abnormal traffic features are generated based on spatiotemporal correlation entropy, and the abnormal traffic features, multi-scale spatial features and time-dependent features are weightedly integrated through the gating weights. The integrated feature data is input into a multi-head self-attention mechanism to determine the cross-spatiotemporal correlation weights between different road sections and / or different time periods to construct a dynamic traffic flow model.

[0034] Specifically, the dynamic traffic change characteristics in the embodiment of the present application reflect the normal periodic and trend changes in traffic flow; abnormal traffic characteristics record special event information such as traffic accidents and emergency control; and the dynamic traffic spatiotemporal graph sequence presents the dynamic evolution of the road network structure and traffic status. These different dimensional information are structured and converted into a multidimensional feature matrix so that it contains both time series characteristics and spatial topological relationships. Specifically, when structuring information of different dimensions, the dynamic traffic change characteristics are first classified according to time period and trend, and periodic statistical values ​​such as flow and speed and trend slopes are extracted; abnormal traffic characteristics are structured according to factors such as event type, occurrence time, and affected road sections; and the dynamic traffic spatiotemporal graph sequence is disassembled into a road network node topological relationship and a state matrix of road sections in each time period. Then, based on the time axis, the corresponding fields of the three types of information are spliced ​​according to the dimensions of time, space, state, and event to form a multidimensional feature matrix. The matrix rows represent time slices, and the columns contain dimensions such as periodic characteristics, trend parameters, spatial topology, and abnormal event codes, thereby realizing information structuring and matrix conversion.

[0035] Through graph convolution operations, the feature information of the nodes is propagated and aggregated on the graph structure to capture the spatial dependencies between the nodes. After stacking multiple layers of graph convolution, multi-scale spatial features from local road sections to global road networks are gradually extracted, such as the traffic capacity of a single intersection, the synergistic effect of adjacent regional road networks, and the macro-topological characteristics of the entire urban traffic network. The embodiment of the present application introduces a traffic spatiotemporal coding mechanism, which generates a unique time position identifier for each traffic period through sine and cosine functions, giving the time series clear timing information. The temporal convolutional network combines these codes to mine the dependencies of traffic flows on the time axis for dynamic traffic change characteristics, such as the traffic growth pattern during the morning rush hour, the periodic changes between weekdays and weekends, etc., thereby obtaining time-dependent features.

[0036] Furthermore, gating weights corresponding to abnormal traffic features are generated based on the previously calculated spatiotemporal correlation entropy. Spatiotemporal correlation entropy measures the uncertainty and correlation of traffic flow in the spatiotemporal dimensions. Higher entropy values ​​indicate a greater likelihood of an anomaly. Gating weights quantify the importance of abnormal traffic features and weight the integration of abnormal traffic features, multi-scale spatial features, and time-dependent features. For high entropy values, abnormal features are given higher weights, making them dominate the overall features and ensuring the model's sensitivity to abnormal events. Under normal circumstances, the contributions of features from various dimensions are balanced to ensure model stability. The integrated feature data is then fed into a multi-head self-attention mechanism. This mechanism calculates attention scores between different road sections and time periods to determine cross-spatiotemporal correlation weights. For example, when analyzing traffic flow changes on a road section during a specific time period, the status of upstream and downstream sections during the same period, as well as historical performance during the same time period, is considered to determine the interactive relationships between traffic flows in the spatiotemporal dimensions and to construct a dynamic traffic flow model.

[0037] Step 102: Through the central node in the traffic network, based on the dynamic traffic flow model, the traffic situation in the future preset time period is deduced and predicted to obtain multiple traffic prediction scenarios, and corresponding traffic control strategies are matched for the multiple traffic prediction scenarios.

[0038] In one implementation of the present application, a federated learning network corresponding to traffic flow data is constructed based on a central node, an edge node, and a dynamic traffic flow model, and a confidence interval prediction network is embedded in the dynamic traffic flow model; wherein, the federated learning network is a hierarchical distributed architecture, and the confidence interval prediction network is used to output the confidence interval of the prediction result. Through the central node, the traffic flow data sent by each edge node in the federated learning network is received, and the weight probability distribution of the confidence interval prediction network is updated through the federated averaging algorithm. The traffic flow data is input into the dynamic traffic flow model to deduce and predict the traffic situation within a preset time period in the future, and a variety of traffic prediction scenarios with confidence intervals are generated through Monte Carlo sampling. Through the central node, the prediction confidence interval of each area is propagated along the road section to the edge node, and when the width of the prediction confidence interval corresponding to any area does not meet the preset threshold, a high-risk scenario identification is triggered.

[0039] Specifically, a hierarchical distributed federated learning network is built based on the central node and edge node architecture in combination with a dynamic traffic flow model. Edge nodes are deployed in various areas of the traffic network, such as intersections and road monitoring points, and are responsible for collecting local traffic flow data. The central node serves as the data and model management hub, coordinating global training and decision-making. At the same time, a confidence interval prediction network is embedded in the dynamic traffic flow model. The network can quantify the uncertainty of the prediction results and output a prediction value containing a confidence interval, such as predicting that the traffic flow on a certain road section will be 500±50 vehicles after 1 hour. Among them, the confidence interval prediction network in the embodiment of the present application can be a Bayesian neural network or a quantile regression model. During training, the input samples are historical traffic flow data and the corresponding time and space context; the output samples are true values ​​and their uncertainty ranges.

[0040] Furthermore, after collecting traffic flow data uploaded by each edge node, the central node uses a federated averaging algorithm to update the weighted probability distribution of the confidence interval prediction network. This algorithm iteratively optimizes the global model by aggregating the local model parameters of edge nodes without transmitting the original data. For example, edge nodes at each intersection train models based on local data, and the central node takes a weighted average of these local model parameters to update the global model, adapting it to the traffic characteristics of different regions. The collected traffic flow data is input into a dynamic traffic flow model and combined with the confidence interval prediction network to predict traffic conditions for a preset future time period. Using Monte Carlo sampling techniques, the model simulates various possible traffic flow evolution paths, such as normal traffic and sudden congestion, and generates multiple prediction scenarios with confidence intervals. For example, when predicting evening rush hour traffic on a particular road section, the model outputs multiple predicted values ​​and their corresponding confidence intervals, such as a traffic range of 800-1000 vehicles at an 80% confidence level. The central node propagates the prediction confidence intervals for each region along the road section, allowing edge nodes to obtain traffic situation prediction information for adjacent areas. If the width of the predicted confidence interval of a certain area exceeds the preset threshold, it is judged as a high-risk scenario, triggering a flag and issuing an early warning.

[0041] In one implementation of the present application, the data of similar traffic scenes in the historical traffic data are divided, and a corresponding control strategy library is established for each scene category. A plurality of traffic prediction scenes within a preset time period in the future are obtained through a dynamic traffic flow model, and the similarity between the plurality of traffic prediction scenes and the historical scene categories is calculated, and the matching scene sets corresponding to each traffic prediction scene are selected based on the similarity. A first timing curve is constructed based on the historical traffic flow data corresponding to each matching scene set, and a second timing curve is constructed based on the traffic data flow corresponding to the traffic prediction scene. A candidate scene is selected from the matching scene set based on the amplitude difference between the first timing curve and the second timing curve. A candidate strategy set is extracted from the control strategy library corresponding to each candidate scenario, and the traffic control strategy corresponding to each traffic prediction scenario is screened in the candidate strategy set with the road network delay rate as the target.

[0042] Specifically, cluster analysis is performed on historical traffic data, and data on similar traffic scenarios, such as weekday morning rush hour, large-scale weekend events, and sudden accidents, are divided into different categories. Each category corresponds to a set of typical features, such as time distribution, traffic pattern, and road network congestion level, and an independent control strategy library is established. For example, for weekday morning rush hour scenarios, the strategy library may contain predefined strategies such as extending the green light time on main roads and opening tidal lanes. Based on multiple traffic forecast scenarios within a preset future time period generated by the dynamic traffic flow model, the similarity between each forecast scenario and the historical scenario category is calculated. By comparing feature vectors, such as traffic peak time and congestion diffusion path, the set of historical scenarios that are most similar to the current forecast scenario is screened out.

[0043] Furthermore, for each matching scenario set, a first time-series curve is constructed based on the historical traffic flow data, reflecting the temporal changes in metrics such as volume and speed. Simultaneously, a second time-series curve is constructed based on the traffic data stream for the current forecast scenario. The time axes of the two curves are aligned for intuitive comparison. The amplitude difference between the first and second time-series curves is calculated. If the difference is within a preset threshold, the corresponding historical scenario is selected as a candidate scenario. These candidate scenarios represent the historical cases that are closest to the current forecast. All available strategies are extracted from the control strategy library corresponding to each candidate scenario to form a candidate strategy set. Using the network delay reduction rate as the optimization objective, the optimal control strategy is selected from the candidate strategy set for each traffic forecast scenario. By evaluating the actual effectiveness of each strategy in the historical scenarios, such as average delay reduction and congestion dissipation speed, and combining the characteristics of the current forecast scenario, such as the impact range and duration, the potential improvement in overall network operating efficiency of each strategy is calculated. Ultimately, the strategy combination that maximizes network delay reduction is selected, enabling precise decision-making and implementation of traffic control.

[0044] Step 103: After an abnormal traffic event is detected, the affected area is delineated based on the propagation impact trend of the abnormal traffic event on the spatial axis, with the road section corresponding to the abnormal traffic event as the center.

[0045] In one implementation of the present application, after an abnormal traffic event is detected, traffic information corresponding to the abnormal traffic event is obtained; wherein the traffic information includes at least the type of traffic event, the time of occurrence of the event, and the traffic volume corresponding to the road section to which it belongs. The road section corresponding to the abnormal traffic event is taken as the center, and the road sections directly or indirectly connected to the road section are taken as adjacent nodes to construct a spatial axis. The traffic information and the spatial axis are input into a preset neural network to output the propagation probability and propagation impact attenuation value of the abnormal traffic event propagating to the adjacent nodes. Based on the comparison result between the propagation probability and the preset probability threshold, a reference space area set is determined in the spatial axis, and based on the surrounding land use type corresponding to each road section and the historical event response frequency, the traffic sensitivity of each road section at the time of the traffic event is determined. Based on the propagation impact attenuation value and the traffic sensitivity, the reference space area set is screened to obtain the affected area.

[0046] Specifically, after detecting an abnormal traffic event, key traffic information related to the event is obtained. This information includes the type of traffic event, the specific time of occurrence, and traffic flow data for the road section where the event occurred, both before and after the event. The road section where the abnormal traffic event occurred is considered the core, and all directly or indirectly connected road sections are classified as adjacent nodes. A spatial axis is constructed based on these nodes, visually displaying the topological relationship between the event section and the surrounding road network. For example, when an accident occurs on a road section, not only the ramps and parallel roads upstream and downstream of the accident section are considered as adjacent nodes, but also the secondary roads connected to these sections, thereby constructing the potentially affected road network structure. The collected traffic information and the constructed spatial axis data are input into a pre-trained neural network model. By learning from a large amount of historical abnormal event data, the model can output the propagation patterns of different types of events in the road network. The model outputs two key indicators: the propagation probability of the abnormal traffic event to each adjacent node (for example, the probability of a road section being affected by the event is 80%), and the impact attenuation value when propagating to each node. The neural network model can be a graph attention network. During training, the input samples are the road network spatial structure data and corresponding traffic information under historical abnormal events, as well as context such as event type and occurrence time; the output samples are the propagation probability and impact attenuation value of abnormal events in each adjacent node. By learning from historical data, the model can grasp the propagation patterns and impact levels of different events in the road network.

[0047] Furthermore, the propagation probability output by the neural network is compared with a preset probability threshold. Adjacent nodes and corresponding road sections with propagation probabilities above the threshold are included in the reference spatial region set as the initial identification of potentially affected areas. Simultaneously, the traffic sensitivity of each road section at the time of the current traffic event is calculated, taking into account the land use types surrounding each road section, such as commercial and residential areas, which are prone to attracting pedestrian and vehicle traffic, and the frequency of historical event responses (i.e., the frequency with which the road section has been affected by similar events in the past). Road sections with low propagation impact attenuation values ​​and high traffic sensitivity are determined to be the core areas affected by the event. Road sections with high attenuation values ​​and low sensitivity are excluded from the affected area, thus determining the affected area.

[0048] Step 104: Divide the traffic control strategy into multiple operation units through the central node, and dynamically combine the operation units based on the traffic information corresponding to the abnormal traffic event to generate a customized control plan.

[0049] In one implementation of the present application, a traffic plan that matches an abnormal traffic event is extracted from a pre-generated traffic control strategy through a central node. The traffic plan is divided into multiple operation units through the central node, and the execution conditions, dependencies and conflict rules are marked for each operation unit, and then entered into the traffic knowledge graph. Through the central node, the traffic information corresponding to the abnormal traffic event is semantically matched and filtered with spatiotemporal constraints with the traffic knowledge graph to screen out a set of applicable operation unit candidates. Through the central node, based on a preset objective function and optimization strategy, the operation units in the operation unit candidate set are dynamically combined to generate a customized control plan corresponding to the abnormal traffic event. The central node performs collaborative relationship verification and conflict detection on the customized control plan through the relationship network corresponding to the traffic knowledge graph to dynamically correct the customized control plan.

[0050] Specifically, the central node, serving as the core hub for traffic control and management, selects traffic plans that match the type and scale of the current abnormal traffic event from a pre-generated traffic control strategy library. For example, if a highway accident is detected, a contingency plan for the accident is extracted, including strategies such as vehicle diversion, temporary control, and rescue dispatch. The central node further breaks down the selected traffic plan into multiple independently executable action units, such as closing the entrance to the accident section and redirecting vehicles to divert from adjacent ramps. Each action unit is annotated with detailed execution conditions, such as execution when traffic volume exceeds a threshold; dependencies, such as the requirement that upstream road sections must be closed before diversion can be initiated; and conflict rules, such as two diversion routes cannot be activated simultaneously. This information is then entered into the traffic knowledge graph to associate entities such as action units, execution conditions, and road network nodes, as well as the relationships between them.

[0051] Furthermore, the central node semantically matches real-time traffic information about abnormal traffic events, such as event type, occurrence time, affected road sections, and current traffic volume, with the knowledge in the traffic knowledge graph. For example, the central node locates the associated node in the graph based on the location of the road section where the event occurred, and matches the corresponding action units in the emergency plan based on the event type. Simultaneously, it filters out action units that do not conform to the actual situation by incorporating temporal and spatial constraints, such as traffic flow restrictions during the current time period and the road section's capacity. Ultimately, it selects a candidate set of action units that are appropriate for the current event, narrowing the scope of available control strategies. Based on pre-set objective functions, such as minimizing network delays and maximizing vehicle traffic efficiency, and optimization strategies, such as prioritizing traffic flow on critical sections, the central node dynamically combines the units in the candidate set. The optimal combination is calculated by comprehensively considering the execution cost, expected effect, and interrelationships of each unit.

[0052] Specifically, the embodiment of the present application constructs a multi-dimensional objective function of the traffic control scheme through the central node; wherein, the multi-dimensional objective function includes efficiency objectives and resource objectives. Through the central node, the operation unit candidate set is screened at the first level based on the constraint relationship of the traffic knowledge graph, and the operation unit candidate set is screened at the second level according to the urgency score. The screened operation unit candidate set is iteratively selected through the greedy algorithm and the multi-dimensional objective function until the preset termination condition is met to obtain the initial combination scheme; wherein the preset termination condition includes that the objective function improvement value is less than the threshold or the resources are exhausted. The initial combination scheme generated by the greedy algorithm is optimized through the reinforcement learning model in the central node. According to the stage of event development, the greedy algorithm and the reinforcement learning model are dynamically weighted. According to the adjusted weights, the greedy algorithm and the reinforcement learning model are reintegrated to obtain customized control schemes corresponding to different stages of event development.

[0053] Furthermore, the central node, serving as the core decision-making unit for traffic control, first constructs a multi-dimensional objective function to measure the performance of control solutions. This objective function encompasses two key dimensions: efficiency objectives focus on traffic operation efficiency, such as reducing overall network delays, increasing average vehicle speeds, and shortening the duration of abnormal events; resource objectives focus on the rational allocation of resources required for control, such as limiting police deployment, the number of traffic light adjustments, and the number of temporary facilities used. By integrating these two objectives, quantitative criteria are provided for subsequent solution screening and optimization. Based on the constraints recorded in the traffic knowledge graph, such as dependencies between control units, conflict rules, and road section capacity limits, a primary screening of the candidate set of control units is performed, eliminating those that do not meet the basic execution conditions. For example, if the knowledge graph indicates that "closing a road section requires prioritizing upstream diversion," control units that do not meet this condition are eliminated. Subsequently, the candidate units are assigned an urgency score based on the attributes of the abnormal traffic event, and secondary screening is performed based on this score, prioritizing those that handle urgent needs.

[0054] After two levels of screening, the central node iteratively selects the remaining operational units using a greedy algorithm. Each time, the greedy algorithm selects the operational unit that best improves the objective function value under the current state for inclusion in the solution. For example, it prioritizes units that significantly reduce delays and consume fewer resources. The algorithm continues iterating until a preset termination condition is met: first, the objective function improvement falls below a threshold, indicating that the solution is near-optimal; second, resource exhaustion occurs, such as when all available police force and equipment have been deployed, making it impossible to execute new units. The resulting initial combination is a rapid response strategy based on local optimality. The central node uses a pre-configured reinforcement learning model to globally optimize the initial solution generated by the greedy algorithm. The reinforcement learning model simulates the control process, allowing the agent to accumulate experience from different solution selections. Based on execution results, such as actual reductions in network delays and resource efficiency, it receives reward feedback, gradually learning the combination that maximizes long-term benefits. For example, the model may discover that certain operational units, while not significantly improving efficiency in the short term, can avoid secondary congestion in the long term. Consequently, the initial solution is adjusted to correct for the local optimality that may have been caused by the greedy algorithm. The reinforcement learning model in the embodiments of this application can be a deep Q-network. During training, the input samples are the initial control plan generated by the greedy algorithm, the current road network status, and environmental parameters; the output samples are the optimized plan combination and the long-term reward estimate of the corresponding strategy. By interacting with the environment, the model uses the reduction in road network delays and resource efficiency as reward signals, iteratively optimizing the strategy to achieve the global optimal adjustment of the initial plan.

[0055] Furthermore, the embodiments of the present application take into account the dynamic evolution of abnormal traffic events, such as the different stages of accident handling from rescue to obstacle removal to restoration of traffic, and dynamically adjust the weights of the greedy algorithm and the reinforcement learning model according to the stage of development of the event. In the early stages of the event, emphasis is placed on rapid response, and the greedy algorithm is given a higher weight to generate an emergency disposal plan; as the event enters the stable processing stage, the weight of the reinforcement learning model is increased to pursue the global optimal solution. By adjusting the weights in real time and reintegrating the outputs of the two algorithms, a customized control plan that fits the needs of different stages of the event can be generated to ensure the timeliness and effectiveness of traffic control.

[0056] Furthermore, the central node comprehensively verifies the generated customized control plan using the relationship network of the traffic knowledge graph. By analyzing the synergies and potential conflicts between the operating units, the feasibility and effectiveness of the plan are tested. If problems are found, such as two diversion routes that may cause intersection congestion, the combination of operating units or the execution order are adjusted, and the plan is dynamically revised to ensure the final control plan is safe and efficient in actual implementation, minimizing the impact of abnormal events on traffic.

[0057] Step 105: Determine the edge nodes corresponding to the affected areas and the local traffic flow data corresponding to the edge nodes. Each edge node jointly optimizes the customized control scheme based on the local traffic flow data using a distributed optimization algorithm to achieve traffic control management.

[0058] In one implementation of the present application, a central node collects the global road network status, determines the impact weight of each road segment, and sends the impact weight and global optimization target to the corresponding edge nodes. Based on constraints, local traffic flow data, and the global optimization target, the edge nodes iteratively optimize the operating unit parameters in the customized control scheme using a distributed optimization algorithm. Adjacent edge nodes exchange the optimization results of boundary intersections and eliminate parameter conflicts through a negotiation mechanism. The edge nodes report the optimized control scheme to the central node, which verifies the global consistency and sends it for execution.

[0059] Specifically, the central node serves as the backbone of the system, collecting traffic status data of the global road network in real time, and calculating the influence weight of each road section based on the road network topology and historical data to quantify the importance of a certain road section in the global traffic. Subsequently, the central node sends the influence weight and the global optimization goal to each edge node. After receiving the instructions from the central node, the edge node uses a distributed optimization algorithm to optimize the operating unit parameters in the customized control scheme based on the local traffic flow data collected by itself and the preset constraints. For example, if the control scheme includes an operating unit for adjusting the timing of traffic lights at intersections, the edge node will iteratively calculate parameters such as the green light duration and the phase switching order, and align the local optimization goal with the global goal while satisfying the constraints.

[0060] Furthermore, for cross-region boundary intersections, adjacent edge nodes are required to exchange their optimization results. If a parameter conflict arises—for example, if the optimization result of node A prioritizes straight-ahead traffic on main roads, while node B prioritizes left turns—the two parties adjust the parameters through a negotiation mechanism. Based on the impact weights and the global objective, the optimal solution for the conflicting area is recalculated. For example, signal timing is adjusted based on the principle of higher weight for main roads. This ensures that the control strategies in the boundary areas are coordinated and consistent, avoiding global congestion caused by local optimization. After completing local parameter optimization, the edge nodes report the final control plan to the central node. The central node verifies the global consistency of the plan, checking whether the optimization results of each edge node meet global constraints, whether boundary coordination effectively eliminates conflicts, and whether the overall objective function meets expectations. If verification passes, the central node distributes the plan to each execution unit. If there is a global inconsistency, feedback is sent to the relevant edge nodes for re-optimization until a globally feasible control plan is formed and executed.

[0061] Figure 2 This is a schematic diagram of a multi-source big data traffic control management system based on a distributed architecture provided by the embodiment of this application. Figure 2As shown, a multi-source big data traffic control and management system based on a distributed architecture includes: an extraction unit 201, which extracts feature data from historical traffic flow data corresponding to a traffic network based on a time sequence, and constructs a dynamic traffic flow model based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network; a prediction unit 202, which uses a central node in the traffic network to deduce and predict traffic conditions within a preset time period in the future based on the dynamic traffic flow model, obtains multiple traffic prediction scenarios, and matches corresponding traffic control strategies for the multiple traffic prediction scenarios; a division unit 203, which, after detecting an abnormal traffic event, demarcates an affected area based on the propagation impact trend of the abnormal traffic event on the spatial axis, with the road section corresponding to the abnormal traffic event as the center; a combination unit 204, which divides the traffic control strategy into multiple operation units through the central node, and dynamically combines the operation units based on the traffic information corresponding to the abnormal traffic event to generate a customized control plan; an optimization unit 205, which determines the edge nodes corresponding to the affected area and the local traffic flow data corresponding to the edge nodes. Each edge node jointly optimizes the customized control plan based on the local traffic flow data using a distributed optimization algorithm to achieve traffic control management.

[0062] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0063] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. It will be apparent to those skilled in the art that various modifications and variations may be made to the embodiments of the present application. However, such modifications or substitutions do not deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A traffic control management method based on distributed architecture multi-source big data, characterized by: The method comprises: Extracting feature data from historical traffic flow data corresponding to a traffic network based on a time sequence, and constructing a dynamic traffic flow model based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network; Through the central node in the traffic network, based on the dynamic traffic flow model, the traffic situation in a preset time period in the future is deduced and predicted to obtain multiple traffic prediction scenarios, and corresponding traffic control strategies are matched for the multiple traffic prediction scenarios; After an abnormal traffic event is detected, the affected area is delineated based on the propagation impact trend of the abnormal traffic event on the spatial axis, with the road section corresponding to the abnormal traffic event as the center; Divide the traffic control strategy into multiple operation units through the central node, and dynamically combine the operation units based on the traffic information corresponding to the abnormal traffic event to generate a customized control plan; Determining edge nodes corresponding to the affected areas and determining local traffic flow data corresponding to the edge nodes, wherein each edge node jointly optimizes the customized control scheme based on the local traffic flow data using a distributed optimization algorithm to achieve traffic control management; The aforementioned matching of the various traffic prediction scenarios with corresponding traffic control strategies specifically includes: Divide the data of similar traffic scenes in historical traffic data and establish a corresponding control strategy library for each scene category; Obtain multiple traffic prediction scenarios within a preset future time period through a dynamic traffic flow model, calculate similarity between the multiple traffic prediction scenarios and historical scenario categories, and select matching scenario sets corresponding to each of the traffic prediction scenarios based on the similarity; Constructing a first time series curve based on the historical traffic flow data corresponding to each of the matching scenario sets, and constructing a second time series curve based on the traffic data flow corresponding to the traffic prediction scenario; selecting a candidate scene from the set of matching scenes according to an amplitude difference between the first timing curve and the second timing curve; Extracting a candidate strategy set from a control strategy library corresponding to each candidate scenario, and screening the traffic control strategy corresponding to each traffic forecast scenario in the candidate strategy set based on the road network delay rate as a target; After an abnormal traffic event is detected, the affected area is delineated based on the propagation impact trend of the abnormal traffic event on the spatial axis, with the road section corresponding to the abnormal traffic event as the center. Specifically, the affected area is delineated based on the propagation impact trend of the abnormal traffic event on the spatial axis. After detecting an abnormal traffic event, obtaining traffic information corresponding to the abnormal traffic event; wherein the traffic information at least includes the type of traffic event, the time of occurrence of the event, and the traffic volume corresponding to the road section to which it belongs; The spatial axis is constructed by taking the road section corresponding to the abnormal traffic event as the center and the road sections directly or indirectly connected to the road section as adjacent nodes; Inputting the traffic information and the spatial axis into a preset neural network to output the propagation probability and propagation impact attenuation value of the abnormal traffic event to adjacent nodes; determining a reference spatial region set in the spatial axis based on a comparison result between the propagation probability and a preset probability threshold; and, determining the traffic sensitivity of each road section at the time when the traffic event occurs based on the surrounding land use types corresponding to each road section and the historical event response frequency; screening the reference space region set according to the propagation impact attenuation value and the traffic sensitivity to obtain the affected region; The traffic control strategy is divided into a plurality of operation units through the central node, and the operation units are dynamically combined based on the traffic information corresponding to the abnormal traffic event, specifically including: Extracting a traffic plan matching the abnormal traffic event from the pre-generated traffic control strategy through the central node; Divide the traffic plan into multiple operation units through the central node, mark the execution conditions, dependencies and conflict rules for each operation unit, and enter them into the traffic knowledge graph; Through the central node, the traffic information corresponding to the abnormal traffic event is semantically matched with the traffic knowledge graph and filtered with spatiotemporal constraints to select a candidate set of applicable operation units; Through the central node, based on a preset objective function and optimization strategy, the operation units in the operation unit candidate set are dynamically combined to generate a customized control plan corresponding to the abnormal traffic event; The central node performs collaborative relationship verification and conflict detection on the customized control plan through the relationship network corresponding to the traffic knowledge graph, so as to dynamically modify the customized control plan; The central node dynamically combines the operation units in the operation unit candidate set based on a preset objective function and optimization strategy to generate a customized control plan corresponding to the abnormal traffic event, specifically including: Constructing a multi-dimensional objective function of a traffic control scheme through the central node; wherein the multi-dimensional objective function includes an efficiency objective and a resource objective; Through the central node, the operation unit candidate set is firstly screened based on the constraint relationship of the traffic knowledge graph, and the operation unit candidate set is secondarily screened based on the urgency score; Iteratively selecting the filtered candidate set of operation units using a greedy algorithm and the multi-dimensional objective function until a preset termination condition is met, thereby obtaining an initial combination scheme; wherein the preset termination condition includes an objective function improvement value being less than a threshold or resource exhaustion; Optimizing the initial combination solution generated by the greedy algorithm through the reinforcement learning model in the central node; Dynamically adjust the weights of the greedy algorithm and the reinforcement learning model according to the stage of event development; According to the adjusted weights, the greedy algorithm and the reinforcement learning model are reintegrated to obtain customized control schemes corresponding to different stages of event development.

2. The method for traffic control and management based on distributed architecture multi-source big data according to claim 1 is characterized in that: The method extracts characteristic data from historical traffic flow data corresponding to the traffic network based on the time sequence, and constructs a dynamic traffic flow model based on the extracted characteristic data and the dynamic spatiotemporal graph corresponding to the traffic network, specifically including: Merging the historical traffic flow data based on the timestamps corresponding to the historical traffic flow data and the location information of the traffic network; Extracting key data from the merged historical traffic flow data through a sliding time window, and determining dynamic traffic change characteristics corresponding to the historical traffic flow data through a preset time series convolutional network based on the extracted key data; Determining the spatiotemporal correlation entropy corresponding to the historical traffic flow data, comparing the spatiotemporal correlation entropy with a preset spatiotemporal correlation entropy, and extracting abnormal traffic features based on the comparison result; Constructing a directed weighted graph based on the traffic network, and dynamically updating the traffic flow state attributes of the directed weighted graph to form a dynamic traffic spatiotemporal graph sequence; The dynamic traffic flow model is constructed based on the dynamic traffic change characteristics, the abnormal traffic characteristics and the dynamic traffic spatiotemporal graph sequence.

3. The method for traffic control and management based on distributed architecture multi-source big data according to claim 2 is characterized in that: The constructing of the dynamic traffic flow model based on the dynamic traffic change characteristics, the abnormal traffic characteristics and the dynamic traffic spatiotemporal graph sequence specifically includes: Constructing a multidimensional feature matrix based on the dynamic traffic change characteristics, the abnormal traffic characteristics, and the dynamic traffic spatiotemporal graph sequence; Based on the multidimensional feature matrix, determining the hidden spatial associations between the road network nodes corresponding to the dynamic traffic spatiotemporal graph sequence, and extracting multi-scale spatial features by stacking multiple layers of graph convolution; The multidimensional feature matrix is ​​processed by a temporal convolutional network and traffic spatiotemporal coding to determine the time-dependent features corresponding to the dynamic traffic change features; wherein the traffic spatiotemporal coding is a time position identifier generated for different traffic periods by sine and cosine functions; generating a gating weight corresponding to the abnormal traffic feature based on the spatiotemporal correlation entropy, and weightedly integrating the abnormal traffic feature, the multi-scale spatial feature, and the time-dependent feature through the gating weight; The integrated feature data is input into a multi-head self-attention mechanism to determine the cross-temporal and spatial correlation weights between different road sections and / or different time periods to construct the dynamic traffic flow model.

4. The method for traffic control and management based on distributed architecture multi-source big data according to claim 1 is characterized in that: The traffic situation within a preset time period in the future is deduced and predicted based on the dynamic traffic flow model through the central node in the traffic network to obtain multiple traffic prediction scenarios, including: Based on the central node, the edge node, and the dynamic traffic flow model, a federated learning network corresponding to the traffic flow data is constructed, and a confidence interval prediction network is embedded in the dynamic traffic flow model; wherein the federated learning network is a hierarchical distributed architecture, and the confidence interval prediction network is used to output a confidence interval of a prediction result; Receiving, through the central node, the traffic flow data sent by each edge node in the federated learning network, and updating the weight probability distribution of the confidence interval prediction network through a federated averaging algorithm; Inputting the traffic flow data into the dynamic traffic flow model to deduce and predict the traffic situation within a preset time period in the future, and generating multiple traffic prediction scenarios with confidence intervals through Monte Carlo sampling; The prediction confidence intervals of each area are propagated to the edge nodes along the road section through the central node, and when the width of the prediction confidence interval corresponding to any area does not meet the preset threshold, a high-risk scenario identification is triggered.

5. The method for traffic control and management based on distributed architecture multi-source big data according to claim 1 is characterized in that: Each edge node jointly optimizes the customized control scheme based on the local traffic flow data using a distributed optimization algorithm, specifically including: The central node collects the global road network status, determines the influence weight of each road segment, and sends the influence weight and the global optimization target to the corresponding edge node; The edge node iteratively optimizes the operating unit parameters in the customized control scheme through a distributed optimization algorithm based on the constraint conditions, local traffic flow data and the global optimization goal; Adjacent edge nodes exchange optimization results of boundary intersections and eliminate parameter conflicts through a negotiation mechanism; The edge node reports the optimized control scheme to the central node, and the central node verifies the global consistency and then issues it for execution.

6. A multi-source big data traffic control and management system based on a distributed architecture, applied to the method according to any one of claims 1 to 5, characterized in that: The system comprises: an extraction unit that extracts feature data from historical traffic flow data corresponding to the traffic network based on a time sequence, and constructs a dynamic traffic flow model based on the extracted feature data and a dynamic spatiotemporal graph corresponding to the traffic network; The prediction unit, through the central node in the traffic network, deduces and predicts the traffic situation within a preset time period in the future based on the dynamic traffic flow model, obtains multiple traffic prediction scenarios, and matches corresponding traffic control strategies for the multiple traffic prediction scenarios, specifically including dividing the data of similar traffic scenarios in the historical traffic data and establishing a corresponding control strategy library for each scenario category; obtaining multiple traffic prediction scenarios within a preset time period in the future through the dynamic traffic flow model, calculating the similarity between the multiple traffic prediction scenarios and the historical scenario categories, and selecting matching scenario sets corresponding to each of the traffic prediction scenarios based on the similarity; constructing a first time series curve based on the historical traffic flow data corresponding to each of the matching scenario sets, and constructing a second time series curve based on the traffic data flow corresponding to the traffic prediction scenario; selecting a candidate scenario from the matching scenario set based on the amplitude difference between the first time series curve and the second time series curve; extracting a candidate strategy set from the control strategy library corresponding to each of the candidate scenarios, and screening the traffic control strategies corresponding to each of the traffic prediction scenarios in the candidate strategy set with the road network delay rate as the target; The division unit, after detecting an abnormal traffic event, takes the road section corresponding to the abnormal traffic event as the center and delineates the affected area according to the propagation impact trend of the abnormal traffic event on the spatial axis, specifically including: obtaining traffic information corresponding to the abnormal traffic event after detecting the abnormal traffic event; wherein the traffic information at least includes the type of traffic event, the time of occurrence of the event, and the traffic volume corresponding to the road section; taking the road section corresponding to the abnormal traffic event as the center and the road sections directly or indirectly connected to the road section as adjacent nodes to construct the spatial axis; inputting the traffic information and the spatial axis into a preset neural network to output the propagation probability of the abnormal traffic event propagating to the adjacent nodes and the propagation impact attenuation value; determining a reference spatial area set in the spatial axis according to a comparison result between the propagation probability and a preset probability threshold; and determining the traffic sensitivity of each road section at the time of the traffic event based on the surrounding land use type corresponding to each road section and the historical event response frequency; screening the reference spatial area set according to the propagation impact attenuation value and the traffic sensitivity to obtain the affected area; The combination unit divides the traffic control strategy into multiple operation units through the central node, and dynamically combines the operation units based on the traffic information corresponding to the abnormal traffic event to generate a customized control plan, which specifically includes: extracting a traffic plan that matches the abnormal traffic event from the pre-generated traffic control strategy through the central node; dividing the traffic plan into multiple operation units through the central node, marking the execution conditions, dependencies and conflict rules for each operation unit, and entering the traffic knowledge graph; performing semantic matching and spatiotemporal constraint filtering on the traffic information corresponding to the abnormal traffic event and the traffic knowledge graph through the central node to screen out a candidate set of applicable operation units; dynamically combining the operation units in the candidate set of operation units based on a preset objective function and optimization strategy to generate a customized control plan corresponding to the abnormal traffic event; performing collaborative relationship verification and conflict detection on the customized control plan through the relationship network corresponding to the traffic knowledge graph to dynamically modify the customized control plan; performing semantic matching and spatiotemporal constraint filtering on the traffic information corresponding to the abnormal traffic event and the traffic knowledge graph through the central node to screen out a candidate set of applicable operation units; Objective function and optimization strategy, dynamically combine the operation units in the operation unit candidate set to generate a customized control plan corresponding to the abnormal traffic event, specifically including: constructing a multi-dimensional objective function of the traffic control plan through the central node; wherein the multi-dimensional objective function includes efficiency objectives and resource objectives; through the central node, the operation unit candidate set is first-level screened based on the constraint relationship of the traffic knowledge graph, and the operation unit candidate set is second-level screened according to the urgency score; the screened operation unit candidate set is iteratively selected through the greedy algorithm and the multi-dimensional objective function until the preset termination condition is met to obtain an initial combination plan; wherein the preset termination condition includes that the objective function improvement value is less than a threshold or resources are exhausted; through the reinforcement learning model in the central node, the initial combination plan generated by the greedy algorithm is optimized; according to the event development stage, the greedy algorithm and the reinforcement learning model are dynamically weighted; according to the adjusted weight, the greedy algorithm and the reinforcement learning model are re-integrated to obtain customized control plans corresponding to different event development stages; The optimization unit determines the edge nodes corresponding to the affected area and the local traffic flow data corresponding to the edge nodes. Based on the local traffic flow data, each edge node jointly optimizes the customized control scheme using a distributed optimization algorithm to achieve traffic control management.

Images