Large-scale urban network multi-agent traffic simulation prediction method, system and equipment

Abstract

The present invention discloses a large-scale urban network multi-agent traffic simulation prediction method, which relates to the technical field of traffic simulation prediction. The present invention includes: constructing a multi-agent urban network model; processing and analyzing traffic demand data; allocating urban network traffic flow; constructing a vehicle driving model; monitoring and providing feedback on urban network traffic; and iterating simulation solutions. The present invention receives travel demand data and processes it, allocating traffic paths to vehicle agents using a classic user equilibrium model. The vehicle agents simulate driving behavior based on a following model and a lane-changing model, and provide data support for the vehicle agents' path selection based on traffic flow macroscopic parameters fed back by the road section agents. Utilizing multi-agent simulation, the simulation accuracy and prediction reliability of traffic flow are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic simulation prediction, and in particular to a method, system and equipment for large-scale urban network multi-agent traffic simulation prediction. Background Art

[0002] The development of urban road traffic is closely linked to the process of urbanization. With population growth and economic development, urban road networks continue to expand, and transportation infrastructure is becoming increasingly well-developed. However, the resulting traffic problems are becoming increasingly prominent, becoming a bottleneck restricting urban development. These urban road traffic problems are primarily manifested in the following aspects: First, traffic congestion is becoming increasingly serious, especially during peak hours in the morning and evening, with vehicles often queuing and moving slowly on major arterial roads and at transportation hubs. Second, traffic accidents are frequent, threatening people's lives and seriously affecting the smooth flow of traffic. Furthermore, with the increase in the number of motor vehicles, environmental pollution is becoming increasingly prominent, with exhaust emissions becoming a significant source of urban air pollution.

[0003] To address these issues, effective traffic control measures are needed. Traditional traffic control relies primarily on manual scheduling and empirical judgment, which is inefficient and difficult to adapt to complex urban traffic environments. Modern traffic control measures, such as Intelligent Transportation Systems (ITS), integrate advanced information technology, communication technology, and control technology to achieve real-time collection, processing, and dissemination of traffic information, thereby improving the level of intelligent traffic control.

[0004] Traffic simulation technology plays a vital role in modern traffic control. By building virtual models of urban roads, traffic simulation can simulate various traffic scenarios, predict traffic flows and conditions, and provide decision support for traffic control. To this end, we propose a large-scale urban network multi-agent traffic simulation prediction method, system, and equipment. Summary of the Invention

[0005] The purpose of the present invention is to provide a large-scale urban network multi-agent traffic simulation and prediction method, system and equipment. By receiving travel demand and processing travel demand data, the classic user equilibrium model is used to allocate traffic paths for vehicle agents. The vehicle agents simulate driving behavior according to the following model and the lane changing model, and provide data support for the vehicle agents to select paths based on the traffic flow macro parameters fed back by the road section agents. The simulation of multiple agents is used to improve the simulation accuracy and prediction reliability of traffic flow.

[0006] According to a first aspect of the present invention, to achieve the above-mentioned purpose, the present invention provides the following technical solution: a large-scale urban network multi-agent traffic simulation prediction method, comprising the following steps:

[0007] S10. Constructing a multi-agent urban network model: Receive urban road network data, create a road network topology based on the road network data, and delineate the simulation area based on the road network topology. Pre-set traffic scene agents, road segment agents, and vehicle agents, establish the connection relationships and interaction logic between the agents, and generate attributes for the large-scale urban network simulation area.

[0008] S20. Traffic Demand Data Processing and Analysis: This model receives multi-source traffic demand data and uses it as input to extract spatiotemporal traffic features. A GCN-LSTM-Attention layered coupled network is used, with the GCN capturing spatial features and the LSTM processing temporal features. The attention mechanism is introduced to enhance the model's ability to identify travel characteristics. By building a neural network model that integrates multiple data sources and deep learning techniques, accurate traffic demand prediction and the generation of an OD matrix are achieved.

[0009] S30. Urban Network Traffic Flow Allocation: Based on traffic demand data, a multi-objective user equilibrium and system optimization allocation model is constructed to allocate traffic flow. The objective function is composed of the integral of the improved road segment resistance function, the system time sum, and the target weight. The additional impedance value caused by users complying with the guidance information is also introduced as a constraint to simulate the impact of traffic guidance on traffic flow distribution.

[0010] S40. Build a vehicle driving model: Combine vehicle physical properties, vehicle spacing, and road conditions to establish a vehicle following model and lane changing model with real-time dynamic parameter updates to achieve dynamic vehicle adjustments;

[0011] S50. Urban Network Traffic Monitoring and Feedback: Based on traffic conditions, the road segment agent calculates traffic parameters and evaluates road traffic conditions, feeding these conditions into the vehicle agent's route selection model. Furthermore, when traffic on a road segment exceeds a capacity threshold, a Bayesian game theory route selection model is constructed to determine the set of paths for each traveler, their type, and their beliefs about other types of travelers. The cost of route selection is calculated based on factors such as travel time and congestion. The benefits are repeatedly calculated and routes are adjusted until the system reaches a Bayesian Nash equilibrium.

[0012] S60. Iterate the simulation scheme, repeat steps S30 to S50, perform traffic simulation of the urban road network, and obtain simulation results.

[0013] Furthermore, the urban road network data includes map data, traffic flow data, and traffic signal configuration data.

[0014] Furthermore, the multi-agent city network model includes the following steps:

[0015] (11) The scope of the large-scale urban road network is consistent with the scope of the simulated urban network. During the simulation process, the urban road network data can be modified. Traffic scenes, road sections, and vehicle data are saved as individual attributes, and the data of different vehicles do not interfere with each other.

[0016] (12) At the start of the simulation, if the vehicle is already within the road network, it will be removed from the simulation until it reaches its destination; if the vehicle is not within the road network at the start of the simulation, the simulation will start from the time it enters the road network and will be removed after it reaches its destination.

[0017] Furthermore, the traffic demand data includes GPS tracks, mobile phone signaling data, city weather information, and date and time data.

[0018] Furthermore, the traffic demand data processing and analysis specifically includes the following steps:

[0019] (21) Receive traffic demand data, pre-process the traffic demand data to eliminate the influence of noise, standardize the data to eliminate the influence of different dimensions, and extract multi-source data features, including speed, route, location, POI stay time, and timestamp;

[0020] (22) Construct a GCN-LSTM-Attention model and construct the traffic network into a graph structure. Nodes represent intersections in the network, and edges represent connections between nodes in time, i.e., road sections. The graph representation uses the adjacency matrix A to represent the connection relationship between nodes and the feature matrix X to represent the time series features of each point. The dimensions are (N, T, F), where N is the number of nodes, T is the number of time steps, and F is the feature dimension. The formula for constructing the GCN layer is as follows:

[0021]

[0022] Where H l is the node representation of the lth layer, represents the normalized adjacency matrix, the initial H (0) =X,W l is the weight matrix of the lth layer, σ is the activation function;

[0023] Constructing LSTM layer to process time series data {h t , c t}=LSTM(X t ,{h t-1 , C t-1}), where X t is the input feature at time step t, h t and c t decibels are the hidden state and cell state with step size t;

[0024] Build the model's Attention mechanism and extract time step features where H is the hidden state at all time steps, α t is the attention weight at time step t;

[0025] (23) The feature data is converted into a sequence form, and the historical data is used to train the model to optimize the model parameters. Based on the prediction results, the traffic OD matrix is ​​generated to represent the traffic flow between different starting points and end points.

[0026] Furthermore, the urban network traffic flow distribution specifically includes the following steps:

[0027] (31) Integrating the elastic demand model into the traffic flow allocation model reflects travelers’ responses to changes in travel costs, including travel time, travel expenses, and travel distance factors;

[0028] Define the demand function D = f(T, C, L), where D is the demand, T is the travel time, C is the travel cost, and L is the travel distance. Use the elasticity coefficient to relate changes in demand to changes in travel time and cost: Where ΔD, ΔT, ΔC, and ΔL represent the changes in demand, travel time, travel cost, and travel distance, respectively. T , η C , η L They represent the elasticity coefficients of travel time, travel cost, and travel distance, respectively, and describe the changes in travel demand under different travel conditions;

[0029] (32) The traffic flow allocation method combining UE and SO is used for initial allocation. The model objective function is the accumulation of the integral of the road resistance function of each road section and the sum of the travel time of all vehicles on the network, which can be expressed as follows:

[0030]

[0031] Where w UE and w SO Represent the weights of user equilibrium and system optimization objectives, x a represents the traffic flow of road section a, t a (w) is the road resistance function of road section a;

[0032] Among them, the road resistance function reflects the situation of traffic congestion and has nonlinear properties. The improved BPR model is adopted:

[0033]

[0034] Where, represents the travel time of road section a in free flow state, C ais the capacity of road section a, x a represents the flow rate of road section a, k a represents the density of road section a, k a，jam represents the congestion density of road section a, 5, β, γ, and δ are all parameters to be calibrated;

[0035] (33) There is an identity constraint on the flow between the path and the OD, that is, on an OD pair, the sum of the flow of all paths is equal to the OD flow, using q rs Denotes the flow of OD pair (r, s), and the flow of the kth path on OD pair (r, s) is recorded as

[0036]

[0037] (34) There is an identity constraint on the flow between the road section and the path, that is, on the road section, the sum of the flow of all the paths passing through the road section is equal to the road section flow, which is expressed as Indicates that the OD pair (r, s) passes through section a on the kth path. If the value is 0, it does not pass through section a. The identity constraint of the section path is expressed as follows:

[0038]

[0039] Where x a represents the flow rate of road section a, is the flow of the kth path on the OD pair;

[0040] (35) The flow of the road section and the flow of the path are non-negative. At the same time, due to the elastic change of travel demand, the flow between OD must be subject to the travel demand, that is:

[0041] x a ≥0, a∈A

[0042]

[0043] 0≤q rs ≤D rs

[0044] Where D rs represents the OD travel demand for (r, s), which is calculated by the travel demand elasticity formula in step (31).

[0045] Furthermore, constructing a vehicle driving model specifically includes the following steps:

[0046] (41) The improved intelligent driver following model is expressed as follows:

[0047]

[0048] Where an (t) represents the acceleration of vehicle n at time t, v n (t) represents the speed of vehicle n at time t, Δv n (t) represents the speed difference between vehicle n and the preceding vehicle at time t, v0 represents the driver’s expected driving speed, represents the driver's desired distance from the vehicle in front, S0 represents the safe distance when the vehicle is stationary, T is the safe time interval, a represents the starting acceleration, and b represents the comfortable deceleration ∈ a ,∈ s They represent the random terms between acceleration and vehicle distance, respectively, and their distributions are calibrated according to the measured data;

[0049] S42: Using the MOBIL lane-changing model, the current vehicle's acceleration gain after changing lanes is added to the acceleration gain of the vehicle behind it in the target lane and the acceleration gain of the vehicle behind it in the current lane, multiplied by the politics coefficient. Lane change is allowed only when the sum exceeds a certain threshold. The model is expressed as follows:

[0050]

[0051] Among them, the acceleration is improved by calculating the intelligent driver following model. Indicates the acceleration gain of the current vehicle after changing lanes, represents the acceleration gain of the vehicle behind in the target lane, It represents the acceleration gain of the vehicle behind in this lane, and p is the politeness coefficient.

[0052] Furthermore, urban network traffic monitoring and feedback specifically include the following steps:

[0053] (51) Define a total of n participants in a traffic trip, and the path selection set S of each traveler i , the type of each trip θ i represents the departure and destination of the travelers, the traveler type space Θ contains all possible combinations of the types of travelers, and each traveler’s belief about other travel types is expressed as P(θ -i |θ i ), that is, traveler i believes that other travelers are of type θ -i probability;

[0054] (52) Traveler i’s benefit function u i (s i , s -i ,θ i ) represents its selection strategy s i and other traveler choice strategies -i The benefit when defining the benefit function u of traveler i i as follows:

[0055] u i (s i , s -i ,θ i )=-(α·travel(s i )+β·congestion(s i , s -i ))

[0056] Among them, α and β are weight parameters used to balance the relative importance of travel time and congestion in the benefit function; travel(s i ) is the path s i The sum of travel time of all road segments, congestion(s i , si) is the path s i The sum of the congestion levels of all road segments; the travel time function and congestion function are expressed as follows:

[0057]

[0058] where time e is the passing time of section e, V e is the flow rate of section e, C e Refers to the traffic capacity of road section e, γ and ε are unknown parameters, which can be calibrated according to historical traffic data. is the congestion level of road segment e, which depends on the paths si chosen by all travelers.

[0059] (53) In the Bayesian game, the Bayesian Nash equilibrium is the optimal strategy combination of a traveler given his travel type and his belief about the travel types of other travelers. Formally, the strategy combination is a Bayesian Nash equilibrium if for each traveler i:

[0060]

[0061] That is, considering that other travelers are also using the optimal strategy during the period, the optimal strategy for traveler i is to maximize his expected benefit.

[0062] According to a second aspect of the present invention, a multi-agent traffic simulation prediction system is provided for implementing the above-mentioned large-scale urban network multi-agent traffic simulation prediction method, comprising:

[0063] The first construction module is used to receive urban road network data, create a road network topology based on the road network data, demarcate the simulation area based on the road network topology, pre-set traffic scene agents, road segment agents, and vehicle agents, establish the connection relationship and interaction logic of each agent, and generate the attributes of the large-scale urban network simulation area;

[0064] The processing and analysis module is used to receive multi-source traffic demand data and use it as input to extract traffic spatiotemporal features. n A hierarchical coupled network uses GCN to capture spatial features and LSTM to process temporal features. The attention mechanism is introduced to enhance the model's ability to identify travel characteristics. By building a neural network model that integrates multiple data sources and deep learning techniques, it can accurately predict traffic demand and generate an OD matrix.

[0065] The allocation module is used to construct a multi-objective user equilibrium and system optimization allocation model based on traffic demand data to allocate traffic flow. The objective function is composed of the improved road section resistance function integral, the system time sum, and the target weight. At the same time, the impedance value added by users due to obeying the induction information is introduced into the constraint condition to simulate the impact of traffic induction on traffic flow distribution.

[0066] The second building block is used to establish a car-following model and a lane-changing model with real-time dynamic parameter updates based on vehicle physical properties, vehicle spacing, and road conditions, in order to achieve dynamic vehicle adjustments.

[0067] The traffic monitoring and feedback module is used to calculate traffic parameters and evaluate road traffic conditions based on the road segment agent. The road conditions are then fed back to the vehicle agent's path selection model. Furthermore, when the road segment traffic exceeds the capacity threshold, a Bayesian game theory path selection model is constructed to determine the set of paths for travelers, their types, and their beliefs about other types of travelers. The cost of path selection is calculated based on factors such as travel time and congestion. The benefits are repeatedly calculated and the paths are adjusted until the system reaches a Bayesian Nash equilibrium.

[0068] The iterative module is used to iterate the simulation scheme, repeat the above modules, conduct traffic simulation of the urban road network, and obtain simulation results.

[0069] According to the third aspect of the present invention, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, the above-mentioned large-scale urban network multi-agent traffic simulation prediction method is adopted.

[0070] The present invention has at least the following beneficial effects:

[0071] The present invention processes travel demand data by receiving travel demand, allocates traffic paths to vehicle agents using the classic user equilibrium model, simulates driving behavior of vehicle agents based on the following model and the lane-changing model, and provides data support for vehicle agents to select paths based on the traffic flow macro parameters fed back by the road section agents. The simulation of multiple agents is used to improve the simulation accuracy and prediction reliability of traffic flow.

[0072] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 Schematic diagram of the process of the present invention;

[0074] Figure 2 This is a schematic diagram of the city OD visualization in the present invention;

[0075] Figure 3 This is a schematic diagram of the simulation demonstration of the vehicle intelligent agent in the urban road network of the present invention;

[0076] Figure 4 This is a visual diagram of the traffic conditions of the urban road network in the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the accompanying drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the embodiments described are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.

[0078] See also Figures 1-4 The present invention provides a technical solution: a large-scale urban network multi-agent traffic simulation prediction method, comprising the following steps:

[0079] S10. Constructing a multi-agent urban network model: Receive urban road network data, create a road network topology based on the road network data, and delineate the simulation area based on the road network topology. Pre-set traffic scene agents, road segment agents, and vehicle agents, establish the connection relationship and interaction logic of each agent, and generate the attributes of the large-scale urban network simulation area. The details are as follows:

[0080] S11. First, conduct a detailed analysis of the urban road network. This includes collecting map data, traffic flow data, traffic signal configurations, and other factors that may affect traffic flow. Using this data, a detailed road network topology can be created to clearly identify the nodes, road segments, and connections within the network.

[0081] S12. Based on the road network topology, design the simulation area divisions. Each simulation area should represent a portion of the urban road network with similar traffic characteristics, such as a commercial district, residential area, or industrial area. Within each simulation area, pre-configure different types of agents based on actual traffic scenarios, including road segment agents and vehicle agents.

[0082] S13. To establish the connections and interaction logic between agents, it is necessary to define their behavioral rules and decision-making processes, including driving rules and lane-changing behavior. Using the multi-agent system (MAS) framework, we can simulate the autonomous behavior and interactions of these agents. We chose to use the MATsim multi-agent simulation platform to implement multi-agent traffic flow simulation in urban road networks.

[0083] S20. Traffic Demand Data Processing and Analysis: This model receives multi-source traffic demand data and uses it as input to extract spatiotemporal traffic features. A GCN-LSTM-Attention layered coupled network is used, with the GCN capturing spatial features and the LSTM processing temporal features. The attention mechanism is introduced to enhance the model's ability to identify travel characteristics. By building a neural network model that integrates multiple data sources and deep learning techniques, accurate traffic demand prediction and the generation of an OD matrix are achieved. The details are as follows:

[0084] (21) Receive traffic demand data, including map data, traffic flow data, traffic signal configuration, and other factors that may affect traffic flow, pre-process the traffic demand data to eliminate noise and standardize the data to eliminate the influence of different dimensions, and extract multi-source data features, including speed, route, location, POI stay time, timestamp, and other data;

[0085] (22) Construct a GCN-LSTM-Attention model and construct the traffic network into a graph structure. Nodes represent intersections in the network, and edges represent connections between nodes in time, i.e., road sections. The graph representation uses the adjacency matrix A to represent the connection relationship between nodes and the feature matrix X to represent the time series features of each point. The dimensions are (N, T, F), where N is the number of nodes, T is the number of time steps, and F is the feature dimension. The formula for constructing the GCN layer is as follows:

[0086]

[0087] Where H l is the node representation of layer 1, represents the normalized adjacency matrix, the initial H (0) =X,W l is the weight matrix of the first layer, σ is the activation function;

[0088] Constructing LSTM layer to process time series data {h t, c t}=LSTM(X t ,{h t-1 , C t-1}), where X t is the input feature at time step t, h t and c t decibels are the hidden state and cell state with step size t;

[0089] Build the model's Attention mechanism and extract time step features where H is the hidden state at all time steps, α t is the attention weight at time step t;

[0090] (23) Convert the feature data into a sequence form, use the historical data to train the model and optimize the model parameters, and generate the traffic OD matrix to represent the traffic flow between different starting points and end points based on the prediction results;

[0091] S30. Urban Network Traffic Flow Distribution: Based on traffic demand data, a multi-objective user equilibrium and system optimization model is constructed to distribute traffic flow. The objective function is composed of the integral of the improved road segment resistance function, the system time sum, and the target weight. The additional impedance value caused by users complying with the guidance information is introduced as a constraint to simulate the impact of traffic guidance on traffic flow distribution. The details are as follows:

[0092] S31. Integrating elastic demand models into traffic flow allocation models reflects travelers' responses to changes in travel costs, including travel time, travel expenses, and travel distance factors;

[0093] Define the demand function D = f(T, C, L), where D is the demand, T is the travel time, C is the travel cost, and L is the travel distance. Use the elasticity coefficient to relate changes in demand to changes in travel time and cost: Where ΔD, ΔT, ΔC, and ΔL represent the changes in demand, travel time, travel cost, and travel distance, respectively. T , η C , η L They represent the elasticity coefficients of travel time, travel cost, and travel distance, respectively, and describe the changes in travel demand under different travel conditions;

[0094] S32. The traffic flow allocation method combining UE and SO is used for initial allocation. The model objective function is the accumulation of the integral of the road resistance function of each road section and the sum of the travel time of all vehicles on the network, which can be expressed as follows:

[0095]

[0096] Where w UE and w SO Represent the weights of user equilibrium and system optimization objectives, x a represents the traffic flow of road section a, t a (w) is the road resistance function of road section a;

[0097] Among them, the road resistance function reflects the situation of traffic congestion and has nonlinear properties. The improved BPR model is adopted:

[0098]

[0099] Where, represents the travel time of road section a in free flow state, C a is the capacity of road section a, x a represents the flow rate of road section a, k a represents the density of road section a, k a，jam represents the congestion density of road section a, 5, β, γ, and δ are all parameters to be calibrated;

[0100] S33. There is an identity constraint on the flow between the path and the OD, that is, on an OD pair, the sum of the flow of all paths is equal to the OD flow, using q rs Denotes the flow of OD pair (r, s), and the flow of the kth path on OD pair (r, s) is recorded as

[0101]

[0102] S34. There is an identity constraint on the flow between the road section and the path, that is, on the road section, the sum of the flow of all paths passing through the road section is equal to the road section flow, which is expressed as Indicates that the OD pair (r, s) passes through section a on the kth path. If the value is 0, it does not pass through section a. The identity constraint of the section path is expressed as follows:

[0103]

[0104] Where x a represents the flow rate of road section a, is the flow of the kth path on the OD pair;

[0105] S35. The flow of the road segment and the flow of the path are non-negative. At the same time, due to the elastic changes in travel demand, the flow between ODs must be subject to the constraints of travel demand, that is:

[0106] x a ≥0, a∈A

[0107]

[0108] 0≤q rs ≤D rs

[0109] Where D rs represents the OD travel demand for (r, s), which is calculated by the travel demand elasticity formula in step S31;

[0110] S40. Build a vehicle driving model: Combine vehicle physical properties, vehicle spacing, and road conditions to establish a vehicle following model and lane changing model with real-time dynamic parameter updates to achieve dynamic vehicle adjustments, as follows:

[0111] S41. The improved intelligent driver following model is described as follows:

[0112]

[0113] Where a n (t) represents the acceleration of vehicle n at time t, v n (t) represents the speed of vehicle n at time t, Δv n (t) represents the speed difference between vehicle n and the preceding vehicle at time t, v0 represents the driver’s expected driving speed, represents the driver's desired distance from the vehicle in front, S0 represents the safe distance when the vehicle is stationary, T is the safe time interval, a represents the starting acceleration, and b represents the comfortable deceleration ∈ a ,∈ s They represent the random terms between acceleration and vehicle distance, respectively, and their distributions are calibrated according to the measured data;

[0114] S42. Using the MOBIL lane-changing model, the current vehicle's acceleration gain after changing lanes is added to the acceleration gain of the vehicle behind it in the target lane and the acceleration gain of the vehicle behind it in the current lane, multiplied by the politics coefficient. Lane change is allowed only when the sum exceeds a certain threshold. The model is expressed as follows:

[0115]

[0116] Among them, the acceleration is improved by calculating the intelligent driver following model. Indicates the acceleration gain of the current vehicle after changing lanes, represents the acceleration gain of the vehicle behind in the target lane, represents the acceleration gain of the vehicle behind in this lane, and p is the politeness coefficient;

[0117] Calibration of model parameters requires collecting actual vehicle behavior data in different traffic scenarios, including speed, acceleration, vehicle distance, lane-changing behavior, etc. This data is obtained through on-board sensors and traffic monitoring systems.

[0118] Based on the collected data, the parameters in the IDM model, such as safety distance, acceleration capability, and deceleration capability, are preliminarily set. Traffic flow is simulated using tools such as SUMO, and the simulation results are compared with actual data to evaluate the model's prediction accuracy.

[0119] The MOBIL lane-changing model focuses on the driver's decision-making parameters, including lane-changing intention, lane-changing timing, and safe distance during the lane-changing process. Similarly, lane-changing behavior is simulated using SUMO and the results are compared with actual lane-changing behavior data to evaluate the model's effectiveness.

[0120] After the model parameters are calibrated, the simulation analysis results are generated as follows Figure 2 ;

[0121] S50. Urban Network Traffic Monitoring and Feedback: Based on traffic conditions, the road segment agent calculates traffic parameters and evaluates road traffic conditions. This information is then fed back to the vehicle agent's path selection model. Furthermore, when traffic on a road segment exceeds the capacity threshold, a Bayesian game theory path selection model is constructed to determine the set of paths for each traveler, their type, and their beliefs about other types of travelers. The cost of path selection is calculated based on factors such as travel time and congestion. The benefits are repeatedly calculated and the paths are adjusted until the system reaches a Bayesian Nash equilibrium, as follows:

[0122] (51) Define a total of n participants in a traffic trip, and the path selection set S of each traveler i , the type of each trip θ i represents the departure and destination of the travelers, the traveler type space Θ contains all possible combinations of the types of travelers, and each traveler’s belief about other travel types is expressed as P(θ -i |θ i ), that is, traveler i believes that other travelers are of type θ -i probability;

[0123] (52) Traveler i’s benefit function u i (s i , s -i ,θ i ) represents its selection strategy s i and other traveler choice strategies -i The benefit when defining the benefit function u of traveler i i as follows:

[0124] u i (s i , s -i ,θ i )=-(α·travel(s i)+β·congestion(s i , s -i ))

[0125] Among them, α and β are weight parameters used to balance the relative importance of travel time and congestion in the benefit function; travel(s i ) is the path s i The sum of travel time of all road segments, congestion(s i , s -i ) is the path s i The sum of the congestion levels of all road segments; the travel time function and congestion function are expressed as follows:

[0126]

[0127] where time e is the passing time of section e, V e is the flow rate of section e, C e Refers to the traffic capacity of road section e, γ and ε are unknown parameters, which can be calibrated according to historical traffic data. is the congestion level of road segment e, which depends on the paths si chosen by all travelers.

[0128] (53) In the Bayesian game, the Bayesian Nash equilibrium is the optimal strategy combination of a traveler given his travel type and his belief about the travel types of other travelers. Formally, the strategy combination is a Bayesian Nash equilibrium if for each traveler i:

[0129]

[0130] That is, considering that other travelers are also using the optimal strategy during the period, the optimal strategy for traveler i is to maximize his expected benefit;

[0131] The road condition is fed back to the vehicle agent, and the vehicle agent reselects the travel path. The process returns to step S30 and iterates the simulation results again. The road section agent detects the road traffic condition. Figure 3 ;

[0132] S60. Iterate the simulation scheme, repeat steps S30 to S50, perform traffic simulation of the urban road network, and obtain simulation results.

[0133] To summarize, this embodiment establishes a multi-agent urban road network simulation model, which includes road section agents and vehicle agents; then, the city's traffic demand information is input, and a path is selected on the road network according to the traffic behavior allocated by the urban road network. The driving behavior of the vehicle agent on the urban road is simulated according to the vehicle following, lane changing, and lane selection models; by monitoring indicators such as the number and speed of road agents, the macro traffic status indicators of the road section are statistically analyzed, such as the average speed of the road section, the road section density, the vehicle flow rate, etc.; finally, the path selection behavior is affected by the traffic status of the road section, and the iteration of the urban road network traffic behavior simulation is realized; through the above-mentioned multi-agent-based traffic simulation prediction method, the prediction ability of the urban road network traffic status can be effectively improved, providing technical support for intelligent transportation systems and urban traffic management.

[0134] Example 2:

[0135] According to a second aspect of the present invention, a multi-agent traffic simulation prediction system is provided for implementing the above-mentioned large-scale urban network multi-agent traffic simulation prediction method, comprising:

[0136] The first construction module is used to receive urban road network data, create a road network topology based on the road network data, demarcate the simulation area based on the road network topology, pre-set traffic scene agents, road segment agents, and vehicle agents, establish the connection relationship and interaction logic of each agent, and generate the attributes of the large-scale urban network simulation area;

[0137] The processing and analysis module receives multi-source traffic demand data and uses it as input to extract spatiotemporal traffic features. It uses a GCN-LSTM-Attention layered coupling network, with GCN capturing spatial features and LSTM processing temporal features. The attention mechanism is introduced to enhance the model's ability to identify travel characteristics. By building a neural network model that integrates multiple data sources and deep learning techniques, it achieves accurate traffic demand prediction and generates an OD matrix.

[0138] The allocation module is used to construct a multi-objective user equilibrium and system optimization allocation model based on traffic demand data to allocate traffic flow. The objective function is composed of the improved road section resistance function integral, the system time sum, and the target weight. At the same time, the impedance value added by users due to obeying the induction information is introduced into the constraint condition to simulate the impact of traffic induction on traffic flow distribution.

[0139] The second building block is used to establish a car-following model and a lane-changing model with real-time dynamic parameter updates based on vehicle physical properties, vehicle spacing, and road conditions, in order to achieve dynamic vehicle adjustments.

[0140] The traffic monitoring and feedback module is used to calculate traffic parameters and evaluate road traffic conditions based on the road segment agent. The road conditions are then fed back to the vehicle agent's path selection model. Furthermore, when the road segment traffic exceeds the capacity threshold, a Bayesian game theory path selection model is constructed to determine the set of paths for travelers, their types, and their beliefs about other types of travelers. The cost of path selection is calculated based on factors such as travel time and congestion. The benefits are repeatedly calculated and the paths are adjusted until the system reaches a Bayesian Nash equilibrium.

[0141] The iterative module is used to iterate the simulation scheme, repeat the above modules, conduct traffic simulation of the urban road network, and obtain simulation results.

[0142] Specifically, the above-mentioned first building module, processing and analysis module, allocation module, second building module, traffic monitoring and feedback module and iteration module can be embedded in a computer processing system. The computer calls the above-mentioned modules to complete the task of traffic simulation prediction based on the above-mentioned large-scale urban network multi-agent traffic simulation prediction method; the above-mentioned first building module, processing and analysis module, allocation module, second building module, traffic monitoring and feedback module and iteration module can perform operations according to the specific steps given in the large-scale urban network multi-agent traffic simulation prediction method.

[0143] It should be noted that it should be understood that the division of the various modules of the above system is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. Moreover, these modules can all be implemented in the form of software called by processing elements; they can also all be implemented in the form of hardware; some modules can also be implemented in the form of processing elements calling software, and some modules can be implemented in the form of hardware. For example, the allocation module can be a separately established processing element, or it can be integrated into a chip of the above-mentioned device. In addition, it can also be stored in the memory of the above-mentioned device in the form of program code, and called and executed by a processing element of the above-mentioned device to perform the functions of the above-mentioned signal processing module. The implementation of other modules is similar. In addition, these modules can all or partly be integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed by the hardware integrated logic circuit in the processor element or software instructions.

[0144] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented by scheduling program code through a processing element, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0145] Example 3:

[0146] The present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores a computer program capable of running on the processor. When the processor loads and executes the computer program, it adopts the large-scale urban network multi-agent traffic simulation prediction method in Example 1.

[0147] It should be noted that the terminal device can be a computer device such as a desktop computer, a laptop computer or a cloud server, and the terminal device includes but is not limited to a processor and a memory. For example, the terminal device can also include input and output devices, network access devices and buses, etc.

[0148] Furthermore, the processor may adopt a central processing unit (CPU). Of course, depending on the actual usage, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. may also be adopted. The general-purpose processor may adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0149] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0150] For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on", "installed on", "fixed on" or "set on" another element, it can be directly on the other element or there can be a central element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there can be a central element at the same time. The terms "vertical", "horizontal", "up", "down", "left", "right" and similar expressions used herein are for illustrative purposes only and are not intended to be the only embodiment.

[0151] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

[0152] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Claims

1. A large-scale urban network multi-agent traffic simulation prediction method, characterized by: The following steps are involved: S10. Constructing a multi-agent urban network model: Receive urban road network data, create a road network topology based on the road network data, and delineate the simulation area based on the road network topology. Pre-set traffic scene agents, road segment agents, and vehicle agents, establish the connection relationships and interaction logic between the agents, and generate attributes for the large-scale urban network simulation area. S20. Traffic Demand Data Processing and Analysis: This model receives multi-source traffic demand data and uses it as input to extract spatiotemporal traffic features. A GCN-LSTM-Attention layered coupled network is used, with the GCN capturing spatial features and the LSTM processing temporal features. The attention mechanism is introduced to enhance the model's ability to identify travel characteristics. By building a neural network model that integrates multiple data sources and deep learning techniques, accurate traffic demand prediction and the generation of an OD matrix are achieved. S30. Urban Network Traffic Flow Allocation: Based on traffic demand data, a multi-objective user equilibrium and system optimization allocation model is constructed to allocate traffic flow. The objective function is composed of the integral of the improved road segment resistance function, the system time sum, and the target weight. The additional impedance value caused by users complying with the guidance information is also introduced as a constraint to simulate the impact of traffic guidance on traffic flow distribution. S40. Build a vehicle driving model: Combine vehicle physical properties, vehicle spacing, and road conditions to establish a vehicle following model and lane changing model with real-time dynamic parameter updates to achieve dynamic vehicle adjustments; S50. Urban Network Traffic Monitoring and Feedback: Based on traffic conditions, the road segment agent calculates traffic parameters and evaluates road traffic conditions. This information is then fed back to the vehicle agent's route selection model. Furthermore, when traffic on a road segment exceeds a capacity threshold, a Bayesian game theory route selection model is constructed to determine the set of paths for each traveler, their type, and their beliefs about other types of travelers. The cost of route selection is calculated based on travel time and congestion factors. The benefits are repeatedly calculated and routes are adjusted until the system reaches a Bayesian Nash equilibrium. S60. Iterate the simulation scheme, repeat steps S30 to S50, perform traffic simulation of the urban road network, and obtain simulation results.

2. The large-scale urban network multi-agent traffic simulation prediction method according to claim 1 is characterized by: The urban road network data includes map data, traffic flow data, and traffic signal configuration data.

3. The large-scale urban network multi-agent traffic simulation prediction method according to claim 2 is characterized by: The multi-agent city network model includes the following steps: (11) The scope of the large-scale urban road network is consistent with the scope of the simulated urban network. During the simulation process, the urban road network data can be modified. Traffic scenes, road sections and vehicle data are saved as individual attributes, and the data of different vehicles do not interfere with each other; (12) At the start of the simulation, if the vehicle is already within the road network, it will be removed from the simulation until it reaches its destination; if the vehicle is not within the road network at the start of the simulation, the simulation will start from the time it enters the road network and will be removed from the simulation until it reaches its destination.

4. The large-scale urban network multi-agent traffic simulation prediction method according to claim 3 is characterized by: The traffic demand data includes GPS tracks, mobile phone signaling data, city weather information, and date and time data.

5. The large-scale urban network multi-agent traffic simulation prediction method according to claim 4 is characterized by: Traffic demand data processing and analysis specifically includes the following steps: (21) Receive traffic demand data, pre-process the traffic demand data to eliminate the influence of noise, standardize the data to eliminate the influence of different dimensions, and extract multi-source data features, including speed, route, location, POI stay time, and timestamp; (22) Construct a GCN-LSTM-Attention model and construct the traffic network into a graph structure. Nodes represent intersections in the network, and edges represent connections between nodes in time, i.e., road sections. The graph representation uses the adjacency matrix A to represent the connection relationship between nodes and the feature matrix X to represent the time series features of each point. The dimensions are (N, T, F), where N is the number of nodes, T is the number of time steps, and F is the feature dimension. The formula for constructing the GCN layer is as follows: In the formula is the node representation of the lth layer, Represents the normalized adjacency matrix, initially , is the weight matrix of the lth layer, σ is the activation function; Building an LSTM layer to process time series data ,in is the input feature at time step t, and are the hidden state and unit state of step size t respectively; Build the model's Attention mechanism and extract time step features , where H is the hidden state at all time steps, is the attention weight at time step t; (23) The feature data is converted into a sequence form, and the model parameters are optimized using the historical data training model. Based on the prediction results, the traffic OD matrix is ​​generated to represent the traffic flow between different starting points and end points.

6. The large-scale urban network multi-agent traffic simulation prediction method according to claim 1 is characterized by: Urban network traffic flow distribution includes the following steps: (31) Incorporating the elastic demand model into the traffic flow allocation model reflects travelers’ responses to changes in travel costs, including travel time, travel expenses, and travel distance factors; Defining the demand function , where D is the quantity demanded, T is the travel time, C is the travel cost, and L is the trip distance, use the elasticity coefficient to relate changes in demand to changes in travel time and cost: ,in , , , , respectively represent the changes in demand, travel time, travel cost, and travel distance, , , They represent the elasticity coefficients of travel time, travel cost, and travel distance, respectively, and describe the changes in travel demand under different travel conditions; (32) The traffic flow allocation method combining UE and SO is used for initial allocation. The model objective function is the accumulation of the integral of the road resistance function of each road section and the sum of the travel time of all vehicles on the network, which can be written as follows: In the formula and Represent the weights of user equilibrium and system optimization objectives respectively, Indicates road section traffic flow, It's a road section The road resistance function; Among them, the road resistance function reflects the situation of traffic congestion and has nonlinear properties. The improved BPR model is adopted: Where, Indicates road section Travel time in free flow conditions, It refers to the road section The traffic capacity, Indicates road section of traffic, Indicates road section The density, Indicates road section The crowding density, These are all parameters to be calibrated; (33) There is an identity constraint on the flow between the path and the OD, that is, on an OD pair, the sum of the flow of all paths is equal to the OD flow. Indicates OD pair The flow rate of OD The flow of the kth path is recorded as : ； (34) There is an identity constraint on the flow between the road section and the path, that is, on a road section, the sum of the flow of all the paths passing through the road section is equal to the road section flow, which is expressed as Indicates OD pair The kth path passes through the road segment If the value is 0, the road section will not be passed. , the identity constraint of the segment path is expressed as follows: Where, Indicates road section of traffic, is the flow of the kth path on the OD pair; (35) The flow of the road section and the flow of the path are non-negative. At the same time, due to the elastic change of travel demand, the flow between OD needs to be restricted by travel demand, that is: Where, Indicates OD pair Travel demand, which is calculated by the travel demand elasticity formula in step (31).

7. The large-scale urban network multi-agent traffic simulation prediction method according to claim 6 is characterized by: Building a vehicle driving model includes the following steps: (41) The improved intelligent driver following model is expressed as follows: Where, Indicates vehicle In time The acceleration when Indicates vehicle In time The speed of Indicates vehicle In time The speed difference with the preceding vehicle, represents the driver's expected driving speed, Indicates the distance the driver expects to maintain from the vehicle ahead. Indicates the safe distance when the vehicle is stationary. is the safety time interval, Indicates the starting acceleration, Indicates comfortable deceleration They represent the random terms between acceleration and vehicle distance, respectively, and their distributions are calibrated according to the measured data; (42) The MOBIL lane-changing model is used. The acceleration gain of the current vehicle after changing lanes is added to the acceleration gain of the vehicle behind the target lane and the acceleration gain of the vehicle behind the current lane, multiplied by the politeness coefficient. When the sum exceeds a certain threshold, lane change is allowed. The model is expressed as follows: Among them, the acceleration is improved by calculating the intelligent driver following model. Indicates the acceleration gain of the current vehicle after changing lanes, represents the acceleration gain of the vehicle behind in the target lane, It represents the acceleration gain of the vehicle behind in this lane, and p is the politeness coefficient.

8. The large-scale urban network multi-agent traffic simulation prediction method according to claim 1 is characterized by: Urban network traffic monitoring and feedback includes the following steps: (51) Define a total of n participants in a transportation trip, and the path selection set of each traveler is , each trip type Represents the departure and destination of travelers, and the type space of travelers Contains all possible types of combinations of travelers, and each traveler's belief about other travel types is expressed as , that is, traveler i believes that other travelers are of type probability; (52) Traveler i’s benefit function Indicates its selection strategy and other traveler choice strategies The benefit when , define the benefit function of traveler i as follows: Among them, α and β are weight parameters used to balance the relative importance of travel time and congestion in the benefit function; Is the path The sum of travel time for all segments on the road, Is the path The sum of the congestion levels of all road segments; the travel time function and congestion function are expressed as follows: in is the passing time of section e, is the flow rate of section e, Refers to the traffic capacity of section e, and They are pending parameters, which can be calibrated based on historical traffic data. is the congestion level of road segment e, which depends on the paths chosen by all travelers ; (53) In the Bayesian game, the Bayesian Nash equilibrium is the best strategy combination of a traveler given his travel type and his belief about the travel types of other travelers. Formally, the strategy combination is a Bayesian Nash equilibrium if for each traveler i: That is, considering that other travelers are also using the optimal strategy during the period, the optimal strategy for traveler i is to maximize his expected benefit.

9. A multi-agent traffic simulation prediction system, used to implement the large-scale urban network multi-agent traffic simulation prediction method according to any one of claims 1 to 8, characterized in that: include: The first construction module is used to receive urban road network data, create a road network topology based on the road network data, demarcate the simulation area based on the road network topology, pre-set traffic scene agents, road segment agents, and vehicle agents, establish the connection relationship and interaction logic of each agent, and generate the attributes of the large-scale urban network simulation area; The processing and analysis module receives multi-source traffic demand data and uses it as input to extract spatiotemporal traffic features. It uses a GCN-LSTM-Attention layered coupling network, with GCN capturing spatial features and LSTM processing temporal features. The attention mechanism is introduced to enhance the model's ability to identify travel characteristics. By building a neural network model that integrates multiple data sources and deep learning techniques, it achieves accurate traffic demand prediction and generates an OD matrix. The allocation module is used to construct a multi-objective user equilibrium and system optimization allocation model based on traffic demand data to allocate traffic flow. The objective function is composed of the improved road section resistance function integral, the system time sum, and the target weight. At the same time, the impedance value added by users due to obeying the induction information is introduced into the constraint condition to simulate the impact of traffic induction on traffic flow distribution. The second building block is used to establish a car-following model and a lane-changing model with real-time dynamic parameter updates based on vehicle physical properties, vehicle spacing, and road conditions, in order to achieve dynamic vehicle adjustments. The traffic monitoring and feedback module is used to calculate traffic parameters and evaluate road traffic conditions based on the road segment agent. The road conditions are then fed back to the vehicle agent's path selection model. Furthermore, when the road segment traffic exceeds the capacity threshold, a Bayesian game theory path selection model is constructed to determine the set of paths for travelers, their types, and their beliefs about other types of travelers. The cost of path selection is calculated based on travel time and congestion factors. The benefits are repeatedly calculated and the paths are adjusted until the system reaches a Bayesian Nash equilibrium. The iterative module is used to iterate the simulation scheme, repeat the above modules, conduct traffic simulation of the urban road network, and obtain simulation results.

10. A terminal device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, the large-scale urban network multi-agent traffic simulation prediction method according to any one of claims 1 to 8 is adopted.

Images