Urban traffic jam intelligent prediction and dispersion system and method thereof

Abstract

The invention relates to the technical field of smart city traffic management, in particular to an urban traffic jam intelligent prediction and dispersion system and a method thereof, and the system integrates a data acquisition module, a traffic prediction module, an event influence evaluation module and an intelligent dispersion module. The data acquisition module updates multi-source traffic data in real time, and a solid foundation is provided for prediction; and the traffic prediction module accurately predicts the traffic flow by using a deep learning model combining a graph neural network and a space-time attention mechanism based on the received data. The event influence evaluation module analyzes the influence of emergencies on traffic and ensures the pertinence of a dispersion strategy; the intelligent dispersion module optimizes a traffic signal lamp control scheme in real time by using deep reinforcement learning in combination with the prediction result and the event influence; the system comprehensively senses the traffic condition through multi-source heterogeneous data fusion, the traffic flow prediction accuracy is remarkably improved, and intelligent and efficient traffic management is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart city traffic management, and in particular to an intelligent prediction and diversion system for urban traffic congestion and a method thereof. Background Art

[0002] With the acceleration of urbanization, urban traffic congestion is becoming increasingly serious, which has brought huge negative impacts on urban residents' travel and urban economic development. Traditional traffic management systems usually use fixed-timing signal light control solutions, which cannot be dynamically adjusted according to real-time traffic conditions, making it difficult to effectively solve the traffic congestion problem.

[0003] The traffic management system in the existing technology has the following main deficiencies: first, the data collection is single, mainly relying on fixed sensors to obtain traffic data, which cannot fully reflect the traffic conditions; second, the traffic flow prediction model is simple, usually using traditional statistical methods or simple machine learning algorithms, and the prediction accuracy is not high; third, there is a lack of assessment and response mechanism for the impact of emergencies. When emergencies such as traffic accidents or bad weather occur, the system cannot respond effectively; finally, the traffic light control solution lacks intelligence and adaptability, and cannot be dynamically optimized and adjusted according to traffic conditions.

[0004] Therefore, there is an urgent need for an urban traffic congestion intelligent prediction and diversion system and method that can comprehensively collect traffic data, accurately predict traffic flow, effectively assess the impact of emergencies and provide intelligent signal light control solutions to solve the problem of urban traffic congestion. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent prediction and diversion system for urban traffic congestion and a method thereof, aiming to solve the problems in the prior art such as single data collection, low accuracy of traffic flow prediction, lack of emergency impact assessment mechanism, and lack of intelligence and adaptability of traffic light control scheme.

[0006] The present invention proposes an intelligent prediction and guidance system for urban traffic congestion, comprising: Data collection module, used to collect and update traffic data of various types and sources in real time; A traffic prediction module, which is in communication with the data acquisition module and is used to receive the traffic data sent by the data acquisition module, build a traffic congestion prediction model based on the traffic data, and obtain a traffic flow prediction result; An event impact assessment module, which is in communication with the traffic prediction module and is used to analyze the impact of emergencies on traffic conditions; The intelligent traffic control module based on deep reinforcement learning is communicated with the traffic prediction module and the event impact assessment module, and is used to provide a real-time traffic light control solution based on the prediction results of the traffic flow and the impact of the emergency on the traffic status.

[0007] Preferably, the traffic data collected by the data collection module includes: Infrastructure data, including vehicle and pedestrian traffic, public transport congestion, road construction congestion, weather impacts and traffic accidents; In-vehicle intelligent terminal data, including vehicle location, vehicle speed and vehicle type; Dynamic sensor data, including traffic light status and pedestrian signal status.

[0008] Preferably, the traffic prediction module comprises: The graph neural network is composed of multiple convolutional layers. It stores historical traffic flow data in a graph structure and gradually extracts the temporal dependency of traffic flow using multiple convolutional layers. Each level of convolution consists of two sub-operations: spatiotemporal graph convolution and spatial graph convolution. The spatiotemporal attention mechanism is used to extract low-level features in the graph neural network, obtain the global dynamic features of the urban road network, and determine the prediction results of traffic congestion based on the global dynamic features.

[0009] Preferably, the graph neural network comprises three graphs: The first figure is a time map constructed based on intersections and time intervals, including historical traffic flow and congestion status; The second figure shows the spatial traffic flow status at each step of the urban road network; The third graph, formed by the dynamic traffic flow in historical data, is used to model the correlation between edges.

[0010] Preferably, the event impact assessment module includes: The event analysis submodule is used to collect and classify historical and real-time events, extract keywords based on event attributes, process and analyze them using natural language processing technology, and generate event-traffic status relationships; The instantaneous impact prediction submodule is used to build an event impact assessment model based on the correlation characteristics between events and urban roads and the traffic flow characteristics of urban roads, and to evaluate the impact of different events at different times on traffic flow.

[0011] Preferably, the intelligent guidance module based on deep reinforcement learning includes: A state assessment submodule, for determining the state of the urban road network based on real-time traffic flow data and the result of the event impact assessment module; The deep reinforcement learning algorithm submodule is used to determine the signal control scheme for different urban road network intersections by interacting with the intelligent traffic control module; The traffic light optimization submodule is used to reduce overall urban road network congestion and travel time by adjusting the traffic light timing plan.

[0012] Preferably, the deep reinforcement learning algorithm submodule includes: A state space construction unit, used to record a set consisting of the current city's traffic flow and corresponding control parameters as a state space; An observation space construction unit is used to record the set of traffic flow observations of all paths in each time slot as the observation space; An action space construction unit for treating a set of control parameters as an action space; The Q value calculation unit is used to construct a graph convolutional neural network as a Q function, which represents the Q value of executing the current action in the current state to go to the next state.

[0013] Preferably, the traffic light optimization submodule is executed by the following steps: Calculate the density of vehicles passing through the intersection, which is defined as the ratio of the number of vehicles predicted to pass through the intersection per unit time to the maximum number of vehicles allowed to pass through the intersection per unit time; Calculate the segment density, which is defined as the density of segments composed of intersections and intersections; Calculate the path speed, defined as the path speed from intersection to intersection; Determine the optimal timing duration and obtain the signal timing of the intersection by optimizing the objective function.

[0014] Preferably, the system further comprises a road network refined modeling module, which is used to: Constructing road segment features, including extracting map feature information such as the coordinates of the start and end points of the road segment, intersection sets, and road segment sets; Construct intersection features, including defining the road set, entrance and exit set, and adjacent intersection set of the intersection, and constructing the intersection state matrix; Perform multi-level congestion assessment, including three-level assessment of intersection congestion and four-level assessment of road section congestion.

[0015] The intelligent prediction and diversion method for urban traffic congestion includes the following steps: Collect and update traffic data of various types and sources in real time through the data collection module; The traffic prediction module receives the traffic data sent by the data acquisition module, builds a traffic congestion prediction model based on the traffic data, and obtains a traffic flow prediction result; Analyze the impact of emergencies on traffic conditions through the event impact assessment module; Through the intelligent traffic control module based on deep reinforcement learning, a real-time traffic light control solution is provided based on the predicted results of the traffic flow and the impact of the emergency on the traffic status.

[0016] The urban traffic congestion intelligent prediction and relief system and method provided by the present invention have the following beneficial effects: 1. Through the fusion of multi-source heterogeneous data, a comprehensive perception of traffic conditions is achieved, providing rich data support for traffic flow prediction; 2. The deep learning model that combines graph neural network with spatiotemporal attention mechanism significantly improves the accuracy of traffic flow prediction; 3. Innovatively introduce an event impact assessment mechanism, which can effectively assess the impact of emergencies on traffic conditions and improve the system's resilience; 4. The intelligent traffic control mechanism based on deep reinforcement learning realizes the adaptive optimization of traffic light control scheme and effectively alleviates traffic congestion; 5. Through refined road network modeling, high-precision expression and congestion assessment of intersections and road sections are achieved, providing an accurate basis for traffic diversion decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is the overall architecture diagram of the urban traffic congestion intelligent prediction and relief system of the present invention; Figure 2 It is a structural diagram of the data acquisition module of the present invention; Figure 3 is a structural diagram of the traffic prediction module of the present invention; Figure 4 is a structural diagram of the event impact assessment module of the present invention; Figure 5 It is a structural diagram of the intelligent guidance module based on deep reinforcement learning of the present invention; Figure 6 It is a structural diagram of the road network refined modeling module of the present invention; Figure 7 It is a flow chart of the method for intelligent prediction and diversion of urban traffic congestion of the present invention; Figure 8 is a structural diagram of the graph neural network of the present invention; Fig. 9 It is a structural diagram of the spatiotemporal attention mechanism of the present invention; Fig.10 It is a flow chart of the deep reinforcement learning algorithm of the present invention. DETAILED DESCRIPTION

[0018] Please refer to the attached Figure 1-10 ,The technical scheme of the present invention is described in detail below in combination with specific ,embodiments.

[0019] Example 1: Overall architecture of the urban traffic congestion intelligent prediction and diversion system: like Figure 1 As shown, the urban traffic congestion intelligent prediction and diversion system provided by the present invention includes a data acquisition module 1, a traffic prediction module 2, an event impact assessment module 3, an intelligent diversion module based on deep reinforcement learning 4 and an optional road network refined modeling module 5.

[0020] The data acquisition module 1 is used to collect and update traffic data of various types and sources in real time. The traffic prediction module 2 is connected in communication with the data acquisition module 1 to receive the traffic data sent by the data acquisition module 1, build a prediction model of traffic congestion based on the traffic data, and obtain the prediction result of traffic flow. The event impact assessment module 3 is connected in communication with the traffic prediction module 2 to analyze the impact of emergencies on traffic conditions. The intelligent traffic control module 4 based on deep reinforcement learning is connected in communication with the traffic prediction module 2 and the event impact assessment module 3 to provide a real-time traffic light control solution based on the prediction results of traffic flow and the impact of emergencies on traffic conditions.

[0021] Data is exchanged between modules through a standard communication interface. Preferably, TCP / IP protocol is used to realize data transmission to ensure the reliability and real-time performance of data transmission. In this embodiment, each module can be deployed on the same server or distributed on different servers to realize data exchange through a network connection.

[0022] Example 2: Specific implementation of the data acquisition module: like Figure 2 As shown, the traffic data collected by the data acquisition module 1 mainly includes three categories: infrastructure data 11, vehicle-mounted intelligent terminal data 12 and dynamic sensor data 13.

[0023] Infrastructure data 11 includes information such as vehicle flow, pedestrian flow, public transportation congestion, road construction congestion, weather impact and traffic accidents. Among them, vehicle flow and pedestrian flow data are usually collected through fixed sensor equipment such as roadside cameras and induction coils, preferably, the data is collected every 30 seconds; public transportation congestion data is obtained through the bus GPS positioning system, preferably, updated every 1 minute; road construction congestion information is obtained through the urban construction management system; weather impact data is obtained through the meteorological department API interface; traffic accident information is pushed in real time through the traffic police system.

[0024] The vehicle-mounted intelligent terminal data 12 includes information such as vehicle location, vehicle speed and vehicle type. These data are mainly collected through vehicle-mounted GPS devices, smartphone navigation applications and dedicated vehicle-mounted terminals. Preferably, the vehicle location and speed data are updated every 10 seconds, and the vehicle type information is recorded when the vehicle is first connected to the system.

[0025] The dynamic sensor data 13 includes information such as the status of traffic lights and pedestrian lights, etc. These data are collected in real time by sensors installed in the signal light control system. Preferably, the data is uploaded immediately when the status of the signal light changes.

[0026] The data acquisition module 1 also includes a data preprocessing unit 14, which is used to clean, standardize and store the collected raw data. Data cleaning mainly includes outlier detection and processing. Preferably, the 3-sigma rule is used to identify outliers, that is, when the data deviates from the mean value by more than 3 times the standard deviation, it is determined to be an outlier; data standardization uses the min-max normalization method to convert the data to the [0,1] interval; data storage uses a distributed database system, preferably, a time series database is used to store real-time data, and a relational database is used to store static data.

[0027] Example 3: Specific implementation of the traffic prediction module: like Figure 3 As shown, the traffic prediction module 2 includes a graph neural network 21 and a spatiotemporal attention mechanism 22. The graph neural network 21 is stacked by multiple levels of convolutional layers. The historical traffic flow data is stored in a graph structure, and the time dependency of the traffic flow is gradually extracted using multiple convolutional layers. Each level of convolution consists of two sub-operations: spatiotemporal graph convolution 211 and spatial graph convolution 212. The spatiotemporal attention mechanism 22 is used to extract low-level features in the graph neural network, obtain the global dynamic features of the urban road network, and determine the prediction results of traffic congestion based on the global dynamic features.

[0028] like Figure 8 As shown, the graph neural network 21 includes three graphs: a first graph 211, a second graph 212 and a third graph 213.

[0029] The first graph 211 is a time graph constructed based on intersections and time intervals, including historical traffic flow and congestion status. minutes (preferably, =5) as nodes and edges of the graph. Each node represents the historical traffic flow. If there are two intersections in the urban road network, the states of the intersections are directly connected through edges. For any two non-adjacent intersections , The distance between , then there is a connecting edge between each two intersections, and the weight of the edge is calculated by the following formula: , in, Indicates road segment The weight of Indicates intersection , The distance and represents a learnable parameter. Preferably, The initial value of is set to 1.5, The initial value of is set to 0.2, and these parameters can be optimized and adjusted through the back-propagation algorithm.

[0030] The second graph 212 represents the spatial traffic flow state of each step of the urban road network. The traffic flow and state of each node in the urban road network are stored in the road network. There are direct connections between adjacent street intersections in the urban road network. Based on the adjacent streets in the urban road network, the directly connected intersections are constructed into subgraphs. , The distance between , there is a connecting edge between every two adjacent intersections, and the edge weight calculation method is the same as the first figure.

[0031] For each subgraph, spatial convolution is used to calculate the dynamic relationship on the spatial graph, as shown in the following formula: , in, Representation Node Features, Representation Node Features, represents the characteristics of node i’s neighbor nodes, represents the set of neighbor nodes of node i, represents the normalization constant, represents the activation function, preferably, the ReLU activation function is adopted.

[0032] The convolution operation is expressed as: , in, represents the learnable parameters, Representation Node and The connection strength between Represents the activation function.

[0033] The third graph 213 is formed by the dynamic traffic flow in the historical data. The historical traffic flow matrix is , the correlation between edges is modeled as follows: , in, and Represents the characteristics of traffic flow at different intersections at different time intervals, Represents the graph neural network in The learnable parameters in the level convolution, Represents the activation function.

[0034] Use spatiotemporal convolution to perform temporal convolution operation, as shown in the following formula: , Among them, it means The dynamic correlation between different intersections in the urban road network at time intervals, represents the learning parameters, represents the activation function. Preferably, The initial value of is set to 0.8 and can be adjusted through the back-propagation algorithm.

[0035] like Fig. 9 As shown, the spatiotemporal attention mechanism 22 calculates the dynamic relationship between different cells using the following formula: , in, It is to predict the traffic flow of the next time interval based on the historical traffic flow. Unet is an attention structure, and the feature set corresponding to the first to nth images is input. The Unet structure adopts an encoder-decoder architecture. The encoder extracts features through convolutional layers and downsampling layers, and the decoder reconstructs features through upsampling layers and convolutional layers, and fuses the features of the encoder and decoder through jump connections. Preferably, both the encoder and decoder use a 3-layer convolutional network, the activation function uses ReLU, and the up- and down-sampling ratio is 2.

[0036] Based on the dynamic relationship, the global dynamic characteristics of the urban road network are obtained, and the prediction results of traffic congestion are determined based on the global dynamic characteristics. Specifically, the global dynamic characteristics are input into the fully connected layer to obtain the traffic flow prediction value of each intersection in the future time period, and then the congestion index is calculated based on the predicted traffic flow and intersection capacity to determine the congestion prediction result. Preferably, the congestion index calculation formula is: Congestion Index , When the congestion index is less than 70%, it is judged to be unobstructed; when the congestion index is between 70% and 90%, it is judged to be slightly congested; when the congestion index is greater than 90%, it is judged to be severely congested.

[0037] Example 4: Specific implementation of the event impact assessment module: like Figure 4 As shown, the event impact assessment module 3 includes an event analysis submodule 31 and an instantaneous impact prediction submodule 32 .

[0038] The event analysis submodule 31 is used to collect and classify historical and real-time events, extract keywords according to event attributes, process and analyze them using natural language processing technology, and generate event-traffic status relationships. Specifically, traffic-related event information is first collected from channels such as traffic police systems, social media, and news reports, and then the events are divided into categories such as traffic accidents, road construction, large-scale activities, and bad weather using a text classification algorithm. Preferably, a BERT-based text classification model is used, and the classification accuracy can reach more than 95%. Next, a named entity recognition algorithm is used to extract key information such as the location, time, and scale of the event. Preferably, a BiLSTM-CRF-based named entity recognition model is used, and the recognition accuracy can reach more than 90%. Finally, based on historical data, a probability association matrix between events and traffic status is established to represent the probability of different types of events affecting different intersections.

[0039] The instantaneous impact prediction submodule 32 is used to construct an event impact assessment model based on the correlation characteristics between events and urban roads and the traffic flow characteristics of urban roads, and to assess the impact of different events on traffic flow at different times. Specifically, the impact of events on different urban roads is modeled as a convolutional neural network, and the convolutional neural network is trained using historical data to predict and assess the impact of each new event on traffic flow: , in, express Time intersection The historical traffic volume, represents the maximum traffic flow when the event occurs, Indicates the minimum traffic flow for the event to occur, Indicates the location where the event occurred; Unet represents a convolutional neural network, and the multiple feature information is combined in each convolutional neural network to obtain an instantaneous impact assessment of each event at the intersection of the urban road network.

[0040] Preferably, the Unet network uses 3 convolution layers, each convolution kernel size is 3×3, the activation function uses ReLU, and the up and down sampling ratio is 2. The input dimension of the convolution network is the feature dimension, and the output dimension is 1, which represents the impact coefficient of the event on traffic flow.

[0041] Based on the instantaneous impact of emergencies on traffic conditions and traffic forecast results, the status assessment results of urban road network intersections are obtained. Specifically, the event impact coefficient is multiplied by the predicted traffic flow to obtain the final traffic flow prediction value after considering the impact of the event: Final traffic flow = predicted traffic flow Event Impact Coefficient Then, based on the final traffic flow and intersection capacity, the congestion index is calculated to determine the final congestion status assessment result.

[0042] Example 5: Specific implementation of the intelligent guidance module based on deep reinforcement learning like Figure 5 As shown, the intelligent traffic control module 4 based on deep reinforcement learning includes a state evaluation submodule 41, a deep reinforcement learning algorithm submodule 42 and a traffic light optimization submodule 43.

[0043] The state assessment submodule 41 is used to determine the state of the urban road network based on the real-time traffic flow data and the results of the event impact assessment module 3. Specifically, let the current time be t, and the traffic flow prediction result obtained by the traffic prediction module 2 is , the event impact assessed by event impact assessment module 3 is , then the overall state can be obtained by the following formula: , Based on the overall state, the input state vector of the intelligent grooming module 4 is determined: , in, Indicates the current status of the city road network; Indicates that according to the time interval Preferably, Fourier series representation is used to capture the intra-day periodic changes of traffic flow: , in, are the Fourier coefficients, is the fundamental frequency, is the number of series terms, preferably, Taking 5 can better fit the daily traffic change pattern.

[0044] like Fig.10 As shown, the deep reinforcement learning algorithm submodule 42 determines the signal light control schemes for different urban road network intersections by interacting with the intelligent diversion module 4. The deep reinforcement learning algorithm submodule 42 includes a state space construction unit 421, an observation space construction unit 422, an action space construction unit 423, and a Q value calculation unit 424.

[0045] The state space construction unit 421 is used to record the set of the current city's traffic flow and the corresponding control parameters as the state space. Specifically, the state space can be expressed as: , in, Indicates The state vector of each intersection includes information such as traffic flow, signal light phase, and number of waiting vehicles.

[0046] The observation space construction unit 422 is used to record the set of traffic flow observations of all paths in each time slot as the observation space. Specifically, the observation space can be expressed as: , in, Indicates The observation vector of each path includes information such as traffic flow, average speed, queue length, etc.

[0047] The action space construction unit 423 is used to regard the control parameter set as an action space. Specifically, the action space can be expressed as: , in, Indicates A signal light control scheme including the duration of each signal light phase.

[0048] The Q value calculation unit 424 is used to construct a graph convolutional neural network as a Q function, which represents the Q value of executing the current action in the current state to go to the next state. Specifically, the Q function can be expressed as: , in, Indicates execution of an action After receiving the instant reward, represents the discount factor, preferably, Taking 0.9 balances immediate rewards and long-term benefits.

[0049] The Q function is implemented through a graph convolutional neural network, and the network structure is as follows: , in, represents the learnable parameters, Representation Node and The connection strength between represents an activation function, preferably, a ReLU activation function is used, preferably, a ReLU activation function is used.

[0050] Preferably, the graph convolutional neural network adopts a 3-layer structure, the number of hidden units in each layer is 128, the learning rate is set to 0.001, the number of training rounds is 1000, and the batch size is 64.

[0051] For the learning of Q value, the temporal difference method is used to update the Q value: , in, represents the learning rate, preferably, Set it to 0.1 to make the value update stable without oscillation.

[0052] In terms of action selection, the ε-greedy strategy is adopted: , in, represents the exploration probability, preferably, The initial value is set to 0.3 and decays linearly to 0.01 during training to balance exploration and exploitation.

[0053] This strategy ensures that the system is sufficiently exploratory in the early stages and can try various signal control schemes. As learning progresses, the system is more inclined to choose the known optimal scheme, thereby improving the stability of the control effect.

[0054] The traffic light optimization submodule 43 adjusts the traffic light timing scheme to reduce overall urban road network congestion and travel time. Specifically, the optimization goal is to minimize the overall travel time and waiting time: , in, Indicates The travel time of each route, represents the waiting time of the i-th path, and is the weight coefficient, preferably, and Both are set to 0.5 to balance the two optimization objectives.

[0055] Traffic light timing optimization is performed through the following steps: First, the intersection vehicle density is calculated, which is defined as the ratio of the number of vehicles predicted to pass through the intersection per unit time to the maximum number of vehicles allowed to pass through the intersection per unit time: , in, Indicates intersection The predicted number of vehicles passing through per unit time, Indicates intersection The maximum number of vehicles allowed to pass per unit time.

[0056] Then calculate the segment density, which is defined as the density of intersections and segments: , Next, the path speed is calculated, which is defined as the path speed from intersection to intersection: , in, Indicates intersection To the intersection The distance Indicates that the intersection and The set of all road segments contained in the path, Indicates road segment Length, Indicates road segment The speed of travel.

[0057] Finally determine the best timing: , in, represents the signal cycle length, preferably, Take 90 seconds, Indicates intersection With intersection Composition of road segments The density of Indicates intersection The number of connected road segments.

[0058] The signal timing of the intersection is obtained by optimizing the objective function: , in, express Time intersection The signal timing, express Time intersection The best time to match Indicates the total duration of signal timing in a time period.

[0059] Example 6: Specific implementation of the road network refined modeling module like Figure 6 As shown, the road network refined modeling module 5 is used to construct road segment features 51 , construct intersection features 52 and perform multi-level congestion assessment 53 .

[0060] Constructing the road segment features 51 includes extracting the starting and ending coordinates of the road segment, the intersection set and the road segment set and other map feature information. Specifically, for each road segment, extract the starting point of the road segment in the map. and end point The latitude and longitude coordinates of the road segment, and map feature information, including the set of all intersections L and the set of road segments E that intersect the road segment, as well as other features related to the road segment. Set the starting point and end point The eigenvector of and , set the feature vector of each intersection in the intersection set L to be the set , the feature vector of each road section in the road section set E is the set .

[0061] Use the road set L and the feature vectors of the intersections in the set to calculate the intersections relative to the starting point and end point Attention: , , in, is the attention matrix, are the eigenvectors of the starting point and the end point, respectively. is the feature vector of the intersection set.

[0062] Constructing intersection features 52 includes defining a set of roads, a set of entrances and exits, and a set of adjacent intersections at the intersection, and constructing an intersection state matrix. Specifically, suppose the intersection The road set is , the export set is , the import collection is , then the intersection The set of adjacent intersections is: , Constructing a set of adjacent intersections The intersection state matrix , the matrix size is the same as the number of adjacent intersections in the intersection set, indicating that the intersection The geometric relationship with its adjacent intersections. If the intersection With intersection If connected, it is 1, otherwise it is 0.

[0063] Construct intersections from a set of adjacent intersections The traffic state matrices of the export and import sets are and , then the set of adjacent intersections of intersection i is Traffic state matrix It can be expressed as: , Constructing intersections Dynamic traffic characteristics , expressed as: , in, For adjacent intersections The traffic state matrix of the exit set, For adjacent intersections The traffic state matrix of the import set, For adjacent intersections traffic flow, For adjacent intersections The number of vehicles passing per unit time, For adjacent intersections Number of intersections passed per unit time Number of vehicles arriving at the exit, For the intersection The number of connected intersections.

[0064] The multi-level congestion evaluation 53 includes a three-level evaluation of intersection congestion and a four-level evaluation of section congestion. Specifically, in terms of intersection congestion evaluation, the congestion level of intersection i is constructed as: , in, For intersection The predicted number of vehicles passing through per unit time, For intersection The maximum number of vehicles allowed to pass per unit time. When Divided into smooth When Classified as light congestion; When Classified as severe congestion.

[0065] In terms of road congestion assessment, the relative congestion level of the road section is used: , in, is the road congestion level, is the traffic demand set of the road section, is the actual number of vehicles passing through the road section, is the number of intersections connected to the road segment. The relative congestion level of road segment i is defined as: , in, For road section The predicted number of vehicles passing through per unit time, For road section The maximum number of vehicles allowed to pass per unit time, For road section With intersection The set of traffic demands at adjacent intersections, For road section Associate intersections in the intersection collection The predicted number of vehicles passing through per unit time.

[0066] Set the road congestion level: The relative congestion level When When When When the traffic is congested, it is set to level 4 (severe congestion).

[0067] Example 7: Implementation of Intelligent Prediction and Diversion Method for Urban Traffic Congestion like Figure 7 As shown, the present invention also provides a method for intelligent prediction and diversion of urban traffic congestion, comprising the following steps: Step 1: collect and update traffic data of various types and sources in real time through the data collection module 1; Step 2: receiving the traffic data sent by the data collection module 1 through the traffic prediction module 2, building a traffic congestion prediction model based on the traffic data, and obtaining a traffic flow prediction result; Step 3: Analyze the impact of the emergency on the traffic status through the event impact assessment module 3; Step 4: Through the intelligent traffic control module 4 based on deep reinforcement learning, a real-time traffic light control solution is provided based on the predicted results of traffic flow and the impact of emergencies on traffic conditions.

[0068] In the implementation process of the method of the present invention, preferably, a distributed computing framework is used to process large-scale traffic data, such as Apache Spark or Apache Flink, to improve the processing power and response speed of the system. At the same time, real-time data stream processing technology is used to ensure that the system can respond to changes in traffic conditions in a timely manner. In addition, in order to improve the reliability and fault tolerance of the system, a microservice architecture is used to design the system so that each functional module can be independently deployed and expanded.

[0069] In specific application scenarios, the present invention can be deployed in the city traffic management center, and by connecting with the existing traffic signal control system, the signal light control plan can be automatically issued and executed. The system can be customized according to actual needs to adapt to the traffic characteristics and management needs of different cities.

[0070] Through the system and method of the present invention, the intelligence level of urban traffic light control can be effectively improved, traffic congestion can be reduced, the urban traffic environment can be improved, road utilization efficiency can be improved, energy consumption and environmental pollution can be reduced, and a more convenient and comfortable travel experience can be provided for citizens.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0072] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Urban traffic congestion intelligent prediction and relief system, characterized by: include: Data collection module, used to collect and update traffic data of various types and sources in real time; A traffic prediction module, which is in communication with the data acquisition module and is used to receive the traffic data sent by the data acquisition module, build a traffic congestion prediction model based on the traffic data, and obtain a traffic flow prediction result; An event impact assessment module, which is in communication with the traffic prediction module and is used to analyze the impact of emergencies on traffic conditions; The intelligent traffic control module based on deep reinforcement learning is communicated with the traffic prediction module and the event impact assessment module, and is used to provide a real-time traffic light control solution based on the prediction results of the traffic flow and the impact of the emergency on the traffic status.

2. The system according to claim 1, characterized in that The traffic data collected by the data collection module includes: Infrastructure data, including vehicle and pedestrian traffic, public transport congestion, road construction congestion, weather impacts and traffic accidents; In-vehicle intelligent terminal data, including vehicle location, vehicle speed and vehicle type; Dynamic sensor data, including traffic light status and pedestrian signal status.

3. The system according to claim 1, characterized in that The traffic prediction module includes: The graph neural network is composed of multiple convolutional layers. It stores historical traffic flow data in a graph structure and gradually extracts the temporal dependency of traffic flow using multiple convolutional layers. Each level of convolution consists of two sub-operations: spatiotemporal graph convolution and spatial graph convolution. The spatiotemporal attention mechanism is used to extract low-level features in the graph neural network, obtain the global dynamic features of the urban road network, and determine the prediction results of traffic congestion based on the global dynamic features.

4. The system according to claim 3, characterized in that The graph neural network contains three graphs: The first figure is a time map constructed based on intersections and time intervals, including historical traffic flow and congestion status; The second figure shows the spatial traffic flow status at each step of the urban road network; The third graph, formed by the dynamic traffic flow in historical data, is used to model the correlation between edges.

5. The system according to claim 1, characterized in that The event impact assessment module includes: The event analysis submodule is used to collect and classify historical and real-time events, extract keywords based on event attributes, process and analyze them using natural language processing technology, and generate event-traffic status relationships; The instantaneous impact prediction submodule is used to build an event impact assessment model based on the correlation characteristics between events and urban roads and the traffic flow characteristics of urban roads, and to evaluate the impact of different events at different times on traffic flow.

6. The system according to claim 1, characterized in that The intelligent guidance module based on deep reinforcement learning includes: A state assessment submodule, for determining the state of the urban road network based on real-time traffic flow data and the result of the event impact assessment module; The deep reinforcement learning algorithm submodule is used to determine the signal control scheme for different urban road network intersections by interacting with the intelligent traffic control module; The traffic light optimization submodule is used to reduce overall urban road network congestion and travel time by adjusting the traffic light timing plan.

7. The system according to claim 6, characterized in that The deep reinforcement learning algorithm submodule includes: A state space construction unit, used to record a set consisting of the current city's traffic flow and corresponding control parameters as a state space; An observation space construction unit is used to record the set of traffic flow observations of all paths in each time slot as the observation space; An action space construction unit for treating a set of control parameters as an action space; The Q value calculation unit is used to construct a graph convolutional neural network as a Q function, which represents the Q value of executing the current action in the current state to go to the next state.

8. The system according to claim 6, characterized in that The traffic light optimization submodule is executed by the following steps: Calculate the density of vehicles passing through the intersection, which is defined as the ratio of the number of vehicles predicted to pass through the intersection per unit time to the maximum number of vehicles allowed to pass through the intersection per unit time; Calculate the segment density, which is defined as the density of segments composed of intersections and intersections; Calculate the path speed, defined as the path speed from intersection to intersection; Determine the optimal timing duration and obtain the signal timing of the intersection by optimizing the objective function.

9. The system according to claim 1, characterized in that The system also includes a road network refined modeling module, which is used to: Constructing road segment features, including extracting map feature information such as the coordinates of the start and end points of the road segment, intersection sets, and road segment sets; Construct intersection features, including defining the road set, entrance and exit set, and adjacent intersection set of the intersection, and constructing the intersection state matrix; Perform multi-level congestion assessment, including three-level assessment of intersection congestion and four-level assessment of road section congestion.

10. A method for intelligently predicting and directing urban traffic congestion, using the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Collect and update traffic data of various types and sources in real time through the data collection module; The traffic prediction module receives the traffic data sent by the data acquisition module, builds a traffic congestion prediction model based on the traffic data, and obtains a traffic flow prediction result; Analyze the impact of emergencies on traffic conditions through the event impact assessment module; Through the intelligent traffic control module based on deep reinforcement learning, a real-time traffic light control solution is provided based on the predicted results of the traffic flow and the impact of the emergency on the traffic status.

Images