Traffic control method, electronic device, and storage medium

Abstract

This application provides a traffic control method, electronic device, and storage medium. The method includes: constructing a hierarchical heterogeneous intelligent agent set corresponding to a target road network, the hierarchical heterogeneous intelligent agent set including a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents; constructing a world model, and issuing a recommended strategy set to the global-level control intelligent agent based on the current traffic state and the traffic control causal graph of the target road network; generating a global control strategy based on the recommended strategy set by the global-level control intelligent agent, and solving the master-slave game equilibrium according to the global control strategy to obtain the target global control strategy; generating multiple intersection control strategies based on the target global control strategy by the multiple regional-level coordination intelligent agents and multiple intersection-level execution intelligent agents, and performing traffic control according to each intersection control strategy to improve traffic control efficiency and effectiveness.

Description

Technical Field

[0001] This application relates to the field of traffic control technology, and more specifically, to a traffic control method, electronic device, and storage medium. Background Technology

[0002] With the continuous growth of urban motor vehicle ownership, traffic congestion has become a core issue restricting urban operational efficiency. Traditional signal control relies on fixed timing or simple adaptive rules, which are insufficient to cope with complex and dynamic changes in traffic flow. Although artificial intelligence technology has been introduced into the field of traffic signal control, existing solutions are still limited to assisting in judgment and cannot achieve full automation of the entire process from perception to decision-making to verification.

[0003] Meanwhile, existing methods generally employ a single-level multi-agent reinforcement learning architecture, where each agent at an intersection only optimizes its local efficiency, easily falling into a collaborative dilemma of local optima and global inefficiency. In addition, they lack the ability to causally infer the causes of congestion, making it impossible to predict the chain reaction across the entire road network caused by policy adjustments. This results in poor policy implementation and may even trigger secondary congestion, making it difficult to adapt to disturbances such as sudden accidents and extreme weather. Summary of the Invention

[0004] The purpose of this application is to address the shortcomings of the prior art by providing a traffic control method, electronic device, and storage medium to solve the practical problem of the limitations of traffic control in the prior art.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a traffic control method, the method comprising: Construct a hierarchical heterogeneous intelligent agent set corresponding to the target road network. The hierarchical heterogeneous intelligent agent set includes a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents. A world model is constructed, and based on the current traffic state and the traffic control causal graph of the target road network, the world model issues a set of recommended strategies to the global-level control agent. The global-level management agent generates a global control strategy based on the recommended strategy set, and solves the master-slave game equilibrium according to the global control strategy to obtain the target global control strategy. The multiple regional-level coordination agents and the multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, and perform traffic control according to each intersection control strategy.

[0006] As an optional implementation, the construction of the world model includes: The world model is trained based on the rules of multiple traffic physics models, the topology data of the target road network, the historical traffic control data of the target road network, and traffic rule data.

[0007] As an optional implementation, the world model, based on the current traffic state and the traffic control causal graph of the target road network, issues a set of recommended strategies to the global-level control agent, including: Based on the current traffic state and the traffic control causal graph, the world model generates multiple candidate traffic control strategies. The pre-built traffic control rule engine verifies each candidate traffic control strategy and determines at least one candidate traffic control strategy to be verified based on the verification results. The simulation results of each candidate traffic control strategy to be verified are determined, the recommended strategy set is generated based on the simulation results, and the recommended strategy set is distributed to the global-level management and control agent.

[0008] As an optional implementation, the generation of a global control policy by the global-level management agent based on the recommended policy set includes: The global-level management and control agent evaluates the global traffic efficiency, secondary congestion risk, and stability of the disturbance scenario of each recommended strategy based on the simulation results of each recommended strategy in the recommended strategy set, and determines the comprehensive score of each recommended strategy. The global-level management agent selects the recommended strategy with the highest comprehensive score as the global control strategy.

[0009] As an optional implementation, the step of solving the master-slave game equilibrium based on the global control strategy to obtain the target global control strategy, and then having the multiple regional-level coordinating agents and the multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, including: The global-level management agent performs a master-slave game equilibrium solution based on the global control strategy and the benefit functions of each layer of the hierarchical heterogeneous multi-agent agent to obtain the target global control strategy, and then distributes the target global control strategy to each regional-level coordination agent. The target global control strategy includes global coordination rules, benefit function constraints, management priority, global objective, timing adjustment boundary and timing constraints. Each regional coordination agent generates a regional coordination strategy based on the target global control strategy, the execution status of the local traffic strategy, and the revenue function of the execution agents at each intersection level under its jurisdiction, and then distributes the strategy to the execution agents at each intersection level under its jurisdiction. Each intersection-level execution agent generates a corresponding intersection control strategy based on the regional coordination strategy and the reward function of the intersection-level execution agent.

[0010] As an optional implementation, the method further includes: The differences between the simulation results at preset time intervals and the measured traffic conditions under the target global control strategy are statistically analyzed. If the difference information meets the preset conditions, traffic scene features are extracted, and the model parameters of the world model are adjusted according to the traffic scene features, the target global control strategy, and the measured traffic state under the target global control strategy to obtain the adjusted world model.

[0011] As an optional implementation, before the world model issues the recommended strategy set to the global-level management agent based on the current traffic state and the traffic control causal graph of the target road network, the following steps are included: The traffic control causal graph is constructed, and the nodes in the traffic control causal graph include traffic scene feature nodes and global control strategy nodes. The edges in the traffic control causal graph are used to represent the causal relationships between the nodes.

[0012] As an optional implementation, the world model generates multiple candidate traffic control strategies based on the current traffic state and the traffic control causal graph, including: The world model extracts the causal feature vector of the current traffic scenario based on the current traffic state, and determines the similarity between the causal feature vector of the current traffic scenario and the feature vector of each historical traffic scenario feature node in the traffic control causal graph. The world model generates multiple candidate traffic control strategies based on their similarity scores.

[0013] Secondly, embodiments of this application provide a traffic control device, the device comprising: The construction module is used to construct a hierarchical heterogeneous intelligent agent set corresponding to the target road network. The hierarchical heterogeneous intelligent agent set includes a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents. The construction module is also used to construct a world model, which, based on the current traffic state and the traffic control causal graph of the target road network, issues a set of recommended strategies to the global-level control agent. The generation module is used to generate a global control strategy based on the recommended strategy set by the global-level management and control agent, and to solve the master-slave game equilibrium according to the global control strategy to obtain the target global control strategy. The multiple regional-level coordination agents and the multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, and to perform traffic control according to each intersection control strategy.

[0014] As an optional implementation, the building module is specifically used for: The world model is trained based on the rules of multiple traffic physics models, the topology data of the target road network, the historical traffic control data of the target road network, and traffic rule data.

[0015] As an optional implementation, the building module is specifically used for: Based on the current traffic state and the traffic control causal graph, the world model generates multiple candidate traffic control strategies. The pre-built traffic control rule engine verifies each candidate traffic control strategy and determines at least one candidate traffic control strategy to be verified based on the verification results. The simulation results of each candidate traffic control strategy to be verified are determined, the recommended strategy set is generated based on the simulation results, and the recommended strategy set is distributed to the global-level management and control agent.

[0016] As an optional implementation, the generation module is specifically used for: The global-level management and control agent evaluates the global traffic efficiency, secondary congestion risk, and stability of the disturbance scenario of each recommended strategy based on the simulation results of each recommended strategy in the recommended strategy set, and determines the comprehensive score of each recommended strategy. The global-level management agent selects the recommended strategy with the highest comprehensive score as the global control strategy.

[0017] As an optional implementation, the generation module is specifically used for: The global-level management agent performs a master-slave game equilibrium solution based on the global control strategy and the benefit functions of each layer of the hierarchical heterogeneous multi-agent agent to obtain the target global control strategy, and then distributes the target global control strategy to each regional-level coordination agent. The target global control strategy includes global coordination rules, benefit function constraints, management priority, global objective, timing adjustment boundary and timing constraints. Each regional coordination agent generates a regional coordination strategy based on the target global control strategy, the execution status of the local traffic strategy, and the revenue function of the execution agents at each intersection level under its jurisdiction, and then distributes the strategy to the execution agents at each intersection level under its jurisdiction. Each intersection-level execution agent generates a corresponding intersection control strategy based on the regional coordination strategy and the reward function of the intersection-level execution agent.

[0018] As an optional implementation, the building module is also used for: The differences between the simulation results at preset time intervals and the measured traffic conditions under the target global control strategy are statistically analyzed. If the difference information meets the preset conditions, traffic scene features are extracted, and the model parameters of the world model are adjusted according to the traffic scene features, the target global control strategy, and the measured traffic state under the target global control strategy to obtain the adjusted world model.

[0019] As an optional implementation, the building module is also used for: The traffic control causal graph is constructed, and the nodes in the traffic control causal graph include traffic scene feature nodes and global control strategy nodes. The edges in the traffic control causal graph are used to represent the causal relationships between the nodes.

[0020] As an optional implementation, the building module is specifically used for: The world model extracts the causal feature vector of the current traffic scenario based on the current traffic state, and determines the similarity between the causal feature vector of the current traffic scenario and the feature vector of each historical traffic scenario feature node in the traffic control causal graph. The world model generates multiple candidate traffic control strategies based on their similarity scores.

[0021] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the traffic control method described in the first aspect above.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the traffic control method described in the first aspect above.

[0023] The beneficial effects of this application are: This application provides a traffic control method, electronic device, and storage medium. It constructs a hierarchical heterogeneous intelligent agent set corresponding to the target road network, including a global-level control agent, multiple regional-level coordination agents, and multiple intersection-level execution agents. A world model is also constructed, which distributes a recommended strategy set to the global-level control agent based on the current traffic state and the traffic control causal graph of the target road network. The global-level control agent generates a global control strategy based on the recommended strategy set and solves the equilibrium through a master-slave game to obtain the target global control strategy. Multiple regional-level coordination agents and multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy and perform traffic control according to each intersection control strategy. By introducing a master-slave game equilibrium solution, the method avoids the collaborative dilemma of local optima and global inefficiency, improving the implementability and effectiveness of the control strategy. Distributing the recommended strategy set to the global-level control agent through the world model improves traffic control efficiency, enabling rapid adaptation to disturbances such as sudden accidents and extreme weather. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram of the architecture of a traffic control system provided in an embodiment of this application; Figure 2 Flowchart of the traffic control method provided in the embodiments of this application Figure 1 ; Figure 3 Flowchart of the traffic control method provided in the embodiments of this application Figure 2 ; Figure 4 Flowchart of the traffic control method provided in the embodiments of this application Figure 3 ; Figure 5 Flowchart of the traffic control method provided in the embodiments of this application Figure 4 ; Figure 6 Flowchart of the traffic control method provided in the embodiments of this application Figure 5 ; Figure 7 Flowchart of the traffic control method provided in the embodiments of this application Figure 6 ; Figure 8 A modular structure diagram of a traffic control device provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0027] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0028] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0029] In the field of traffic control, traditional signal control relies on fixed timing or simple adaptive rules, which are insufficient to cope with complex and dynamic changes in traffic flow. Although artificial intelligence technology has been introduced into traffic control, existing solutions are still limited to assisting in judgment and cannot achieve full-process automation from perception to decision-making to verification. Furthermore, existing methods generally employ a single-level multi-agent reinforcement learning architecture, where each intersection agent only optimizes local efficiency, easily falling into a collaborative dilemma of local optima and global inefficiency. In addition, the lack of causal deduction capabilities for congestion causes makes it impossible to predict the chain reaction across the entire road network resulting from policy adjustments, leading to poor policy implementation and even secondary congestion, and making it difficult to adapt to disturbances such as sudden accidents and extreme weather.

[0030] Based on the above-mentioned problems, this application provides a traffic control method that improves traffic control efficiency and effectiveness by using a hierarchical heterogeneous intelligent agent set and a world model, thereby achieving autonomous evolution of traffic control.

[0031] Figure 1 A schematic diagram of the architecture of the traffic control system provided in the embodiments of this application is shown below. Figure 1 As shown, the traffic control system includes a world model, a signal control rule engine, a fluid dynamics model, and a hierarchical heterogeneous set of agents corresponding to the target road network, consisting of a global-level control agent, multiple regional-level coordination agents, and multiple intersection-level execution agents.

[0032] The system comprises a world model for outputting recommended policy sets to the global-level management agent, a hierarchical heterogeneous agent set for real-time traffic state perception and hierarchical execution of traffic control policies, a signal control rule engine for validating traffic control policies to improve traffic control safety and compliance, and a fluid dynamics model for simulating and extrapolating traffic control policies to enhance their implementability and effectiveness.

[0033] Figure 2 Flowchart of the traffic control method provided in the embodiments of this application Figure 1 The subject executing this method can be any electronic device with computing power. For example... Figure 2 As shown, the method includes: S101. Construct a hierarchical heterogeneous intelligent agent set corresponding to the target road network.

[0034] The hierarchical heterogeneous intelligent agent set includes a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents.

[0035] Optionally, refer to Figure 1 The hierarchical heterogeneous intelligent agent set corresponding to the target road network includes a global-level control intelligent agent, Regional-level coordinated intelligent agents and Each intersection-level execution intelligent agent.

[0036] The decision-making algorithms and reward functions (i.e., optimization objectives) used by each layer of intelligent agents are completely heterogeneous. Specifically, each intersection-level execution agent is deployed at an independent intersection and is responsible for sensing and reporting structured traffic data such as traffic flow, vehicle speed, and queue length in real time. Based on the coordination rules issued by the regional-level coordinating agent, a deep reinforcement learning algorithm is used to adjust the signal timing of the local intersection to ensure the traffic efficiency of the local intersection. The optimization objective is to maximize the traffic efficiency of the local intersection and minimize queuing and delays.

[0037] Each regional-level coordination agent manages a regional road network containing multiple intersections. It uses a causal inference algorithm to achieve arterial coordinated control and congestion root cause localization. It issues regional coordination strategies to the execution agents at each intersection level under its jurisdiction and reports the congestion root causes to the global-level management agent. The optimization goal is to maximize the overall traffic efficiency of the region without any traffic overflow.

[0038] The global-level management agent is responsible for the global management of the entire target road network. It generates a global control strategy based on the recommended strategy set issued by the world model, and solves the master-slave game equilibrium based on the global control strategy. It then issues the target global control strategy to the lower-level regional coordination agents to optimize the goal of maximizing the global traffic efficiency of the target road network.

[0039] S102. Construct a world model, which, based on the current traffic conditions and the causal graph of traffic control in the target road network, distributes a set of recommended strategies to the global-level control agent.

[0040] Optionally, a world model with causal reasoning capabilities can be constructed, which is a world model specifically for the transportation field.

[0041] The world model analyzes the current traffic state of the target road network through structured traffic data or unstructured disturbance information reported by the global-level control agent. Based on the pre-constructed traffic control causal graph of the target road network, it generates a set of recommended strategies and outputs the set of recommended strategies to the global-level control agent for decision-making.

[0042] S103. A global-level management agent generates a global control strategy based on a recommended strategy set, and solves the equilibrium of a master-slave game based on the global control strategy to obtain the target global control strategy. Multiple regional-level coordination agents and multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, and perform traffic control according to each intersection control strategy.

[0043] Optionally, the global-level control agent selects the optimal global control strategy from the recommended strategy set, and solves the master-slave game equilibrium based on the global control strategy to determine the target global control strategy. The target global control strategy can avoid the global deterioration problem caused by local optimization.

[0044] Specifically, in the game, the global-level control agent acts as the leader, while the regional-level coordinating agents and the intersection-level execution agents act as followers. The leader's game objective is the global benefit. Maximizing the regional payoff is the game objective of the regional coordinating agent among the followers. Maximizing, the game objective of the intersection-level executive agent in the follower is local gain. maximize.

[0045] After the global-level control agent determines the target global control strategy, multiple regional-level coordination agents and multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, and perform traffic control at local intersections according to each intersection control strategy, such as adjusting traffic light cycles, phase sequence, and green light duration.

[0046] In this embodiment, a hierarchical heterogeneous intelligent agent set corresponding to the target road network is constructed, including a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents. A world model is also constructed, and the world model issues a recommended strategy set to the global-level control intelligent agent based on the current traffic state and the traffic control causal graph of the target road network. The global-level control intelligent agent generates a global control strategy based on the recommended strategy set, and performs master-slave game equilibrium solution based on the global control strategy to obtain the target global control strategy. The multiple regional-level coordination intelligent agents and multiple intersection-level execution intelligent agents generate multiple intersection control strategies based on the target global control strategy, and perform traffic control according to each intersection control strategy. By introducing master-slave game equilibrium solution, the cooperative dilemma of local optima and global inefficiency is avoided, improving the implementation and effectiveness of the control strategy. The world model issues a recommended strategy set to the global-level control intelligent agent to improve traffic control efficiency and quickly adapt to disturbances such as sudden accidents and extreme weather.

[0047] As an optional implementation, the construction of the world model in step S102 above includes: A world model is trained based on rules from multiple traffic physics models, topological data of the target road network, historical traffic control data of the target road network, and traffic rule data.

[0048] Optionally, the rules of multiple traffic physics models (such as car-following models, lane-changing models, and fluid dynamics models) can be injected into the initial world model through fine-tuning instructions to train the initial world model to learn the evolution of traffic flow, such as the propagation characteristics of traffic flow and the diffusion mechanism of congestion.

[0049] The topological data of the target road network (such as the topological structure of the target road network, intersection connections, lane configuration, turning restrictions, speed limit rules, and other static data) is integrated into the initial world model in the form of knowledge graph embedding, so that the initial world model can grasp the spatial structural characteristics of the target road network.

[0050] Historical traffic control data of the target road network (such as all historical control cases of the target road network in the past 3 years) are integrated into the initial world model through retrieval-augmented generation (RAG) so that the initial world model has the ability to call upon historical experience.

[0051] Traffic rules are organized into structured instruction data, which are then injected into the initial world model through instruction fine-tuning. The initial world model is then trained to learn the traffic rules, resulting in a world model specific to the traffic domain.

[0052] In this embodiment, a world model is trained based on rules from multiple traffic physics models, topological data of the target road network, historical traffic control data of the target road network, and traffic rule data. This enables the traffic-specific world model to learn the evolution of traffic flow and traffic rules, grasp the spatial structural characteristics of the target road network, and possess the ability to invoke historical experience.

[0053] Figure 3 Flowchart of the traffic control method provided in the embodiments of this application Figure 2 ,like Figure 3 As shown, in step S102 above, the world model, based on the current traffic state and the traffic control causal graph of the target road network, issues a set of recommended strategies to the global-level control agent, including: S201. The world model generates multiple candidate traffic control strategies based on the current traffic conditions and the causal graph of traffic control.

[0054] Optionally, the world model retrieves multiple historical traffic control strategies that are relatively well matched to the current traffic state from the traffic control causal graph based on the current traffic state, and uses them as multiple candidate traffic control strategies.

[0055] S202. The pre-built traffic control rule engine verifies each candidate traffic control strategy and determines at least one candidate traffic control strategy to be verified based on the verification results.

[0056] Optionally, the world model outputs each candidate traffic control strategy to a pre-built traffic control rule engine. Before entering the counterfactual simulation stage, the traffic control rule engine performs a pre-verification of each candidate traffic control strategy to obtain the verification results.

[0057] Based on the verification results, the candidate traffic control strategies that pass the verification are selected as candidate traffic control strategies to be verified, thus preventing candidate traffic control strategies that violate traffic rules from entering the counterfactual simulation stage.

[0058] S203. Determine the simulation results of each candidate traffic control strategy to be verified, generate a recommended strategy set based on the simulation results, and distribute the recommended strategy set to the global-level management and control agent.

[0059] Optionally, the world model outputs key parameters required for simulation to the fluid dynamics model based on each candidate traffic control strategy to be verified, such as the sensitivity coefficient of the car-following model, the traffic flow propagation speed, and the congestion diffusion coefficient.

[0060] The fluid dynamics model performs multi-threaded parallel counterfactual simulations for the key parameters of each candidate traffic control strategy to be verified. Each thread independently simulates the global traffic state changes of the target road network over a future period (e.g., 30 minutes) under each candidate traffic control strategy to be verified, and obtains the simulation results of the candidate traffic control strategy to be verified.

[0061] Based on the simulation results of the candidate traffic control strategies to be verified, a recommended strategy set is generated, and the recommended strategy set and its simulation results are sent to the global-level control agent.

[0062] In this embodiment, a world model generates multiple candidate traffic control strategies based on the current traffic state and the traffic control causal graph. A pre-built traffic control rule engine verifies each candidate strategy, and at least one candidate strategy to be verified is determined based on the verification results. The simulation results of each candidate strategy to be verified are determined, and a recommended strategy set is generated based on the simulation results. This recommended strategy set is then distributed to the global-level control agent. This provides the global-level control agent with a relatively accurate recommended strategy set for efficient decision-making.

[0063] Figure 4 Flowchart of the traffic control method provided in the embodiments of this application Figure 3 ,like Figure 4 As shown, in step S103 above, the global-level management agent generates a global control policy based on the recommended policy set, including: S301. The global-level control agent evaluates the global traffic efficiency, secondary congestion risk, and stability of the disturbance scenario of each recommended strategy based on the simulation results of each recommended strategy in the recommended strategy set, and determines the comprehensive score of each recommended strategy.

[0064] Optionally, the global-level control agent evaluates the global traffic efficiency, secondary congestion risk, and stability under disturbance scenarios of each recommended strategy based on the simulation results of each recommended strategy in the recommended strategy set, and obtains the global traffic efficiency score of each recommended strategy. Secondary congestion risk score and stability score in disturbed scenarios .

[0065] Global efficiency score of each recommendation strategy for the global management agent Secondary congestion risk score and stability score in disturbed scenarios Weighted fusion is performed to obtain the comprehensive score of each recommendation strategy. .

[0066] S302, The global-level management agent selects the recommended strategy with the highest comprehensive score as the global control strategy.

[0067] Optionally, the global-level control agent's comprehensive score for each recommendation strategy. Sort the results and calculate the overall score. The highest recommended strategy is used as the global control strategy. This allows for master-slave game theory solutions based on the global control strategy, thereby improving global traffic efficiency and stability in disturbed scenarios, and reducing the risk of secondary congestion.

[0068] In this embodiment, a global-level control agent evaluates the global traffic efficiency, secondary congestion risk, and stability of each recommended strategy based on simulation results from the recommended strategy set. It then determines the comprehensive score of each strategy and selects the strategy with the highest comprehensive score as the global control strategy. This allows for master-slave game theory based on the global control strategy, thereby improving global traffic efficiency, stability of the disturbance scenario, and reducing the risk of secondary congestion.

[0069] Figure 5 Flowchart of the traffic control method provided in the embodiments of this application Figure 4 ,like Figure 5 As shown, in step S103 above, a master-slave game equilibrium is solved based on the global control strategy to obtain the target global control strategy. Multiple regional-level coordinating agents and multiple intersection-level executing agents then generate multiple intersection control strategies based on the target global control strategy, including: S401. The global-level management agent performs a master-slave game equilibrium solution based on the global control strategy and the benefit functions of each layer of the hierarchical heterogeneous multi-agent agent to obtain the target global control strategy, and then distributes the target global control strategy to each regional-level coordination agent.

[0070] The target global control strategy includes global coordination rules, revenue function constraints, control priorities, global objectives, timing adjustment boundaries, and timing constraints.

[0071] Optionally, the global-level management agent acts as the leader, and solves the master-slave game equilibrium based on the global control strategy and the payoff functions of each layer of the hierarchical heterogeneous multi-agent agent. This yields a target global control strategy that includes global coordination rules, payoff function constraints, management priorities, global objectives, timing adjustment boundaries, and timing constraints, which is then distributed to each regional-level coordinating agent.

[0072] Among them, the global-level control agent achieves the intrinsic alignment of the global control objectives of the target road network from the top level, resolving the local optimum problem caused by the conflict of objectives between levels. By issuing global control strategies, it defines clear decision boundaries for each lower-level agent, preventing the lower-level agents from deviating from the global optimum due to autonomous decision-making, and ensuring global benefits. maximize.

[0073] S402. Each regional-level coordinating agent generates a regional coordination strategy based on the target global control strategy, the execution status of the local traffic strategy, and the revenue function of the execution agents at each intersection level under its jurisdiction, and then distributes the strategy to the execution agents at each intersection level under its jurisdiction.

[0074] Optionally, each regional coordination agent uses the global control strategy as a constraint boundary, combines the real-time execution status of the regional traffic strategy with the revenue function of the intersection-level agents under its jurisdiction, generates a coordination strategy adapted to the region, and distributes it to the intersection-level execution agents.

[0075] Based on this, differentiated regional control can be achieved within global constraints, ensuring that regional decisions do not deviate from the overall control objectives, while also enabling dynamic adjustment of trunk green waves and precise location of congestion root causes by combining real-time regional traffic conditions. This balances global consistency with regional flexibility and improves the efficiency of collaborative control of the regional road network.

[0076] S403. Each intersection-level execution agent generates a corresponding intersection control strategy based on the regional coordination strategy and the reward function of the intersection-level execution agent.

[0077] Optionally, each intersection-level execution agent, constrained by the regional coordination strategy issued by the upper layer, makes real-time decisions based on its own reward function to generate the local intersection control strategy.

[0078] Based on this, short-term real-time response at the intersection level can be achieved, quickly adapting to local traffic flow fluctuations and ensuring intersection traffic efficiency. At the same time, all timing adjustments are constrained by regional coordination strategies, effectively avoiding negative phenomena such as traffic overflow and secondary congestion in the surrounding area caused by individual intersection optimization, and achieving a balance between local traffic efficiency and global coordination goals.

[0079] In this embodiment, the global-level management agent performs a master-slave game equilibrium solution based on the global control strategy and the reward functions of each layer of the hierarchical heterogeneous multi-agent system to obtain the target global control strategy, which is then distributed to each regional-level coordination agent. The target global control strategy includes global coordination rules, reward function constraints, control priorities, global objectives, timing adjustment boundaries, and timing constraints. Each regional-level coordination agent generates a regional coordination strategy based on the target global control strategy, the execution status of the local traffic strategy, and the reward functions of the intersection-level execution agents under its jurisdiction, and distributes it to the intersection-level execution agents under its jurisdiction. Each intersection-level execution agent then generates its corresponding intersection control strategy based on the regional coordination strategy and its own reward function. This approach avoids getting trapped in local optima and achieves global optimization.

[0080] Figure 6 Flowchart of the traffic control method provided in the embodiments of this application Figure 5 ,like Figure 6As shown, the method also includes: S501. The difference between the simulation results of the preset time interval and the measured traffic state under the target global control strategy.

[0081] Optionally, the mean absolute percentage error (MAPE) between the simulation results at preset time intervals and the measured traffic conditions under the global control strategy is statistically analyzed, and the MAPE is used as standardized difference information, i.e., simulation deviation value.

[0082] By statistically analyzing the MAPE between the simulation results at preset time intervals and the measured traffic conditions under the global control strategy, the degree of deviation between the simulation results at preset time intervals and the measured traffic conditions under the global control strategy can be accurately quantified.

[0083] S502. If the difference information meets the preset conditions, the traffic scene features are extracted, and the model parameters of the world model are adjusted according to the traffic scene features, the target global control strategy, and the measured traffic state under the target global control strategy to obtain the adjusted world model.

[0084] Optionally, at least one deviation threshold is preset as a condition for triggering the adjustment of the world model. When the statistically obtained difference information meets the preset threshold requirement, the current traffic scene features are automatically extracted. Combined with the target global control strategy issued this time and the measured traffic state under the strategy, the model parameters of the world model are fine-tuned to obtain the adjusted world model.

[0085] The preset conditions can be that the global average simulation deviation value of the target road network is greater than or equal to a first threshold, or that the maximum simulation deviation value of a single intersection is greater than or equal to a second threshold. For example, the first threshold can be 10%, and the second threshold can be 20%.

[0086] The extracted traffic scene features include time-period attributes (such as morning peak, evening peak, etc.), weather conditions, congestion root cause types, disturbance types, and road network traffic distribution. A lightweight fine-tuning method using Low-Rank Adaptation (LoRA) is employed, with the optimization objective of minimizing the mean squared error (MSE) between simulation results and measured data, thus achieving closed-loop self-optimization of the world model.

[0087] In this embodiment, the differences between simulation results at preset time intervals and measured traffic states under the target global control strategy are statistically analyzed. If the differences meet preset conditions, traffic scene features are extracted. Based on the traffic scene features, the target global control strategy, and the measured traffic states under the target global control strategy, the model parameters of the world model are adjusted to obtain the adjusted world model. This achieves closed-loop self-optimization of the world model, gradually improving the strategy recommendation accuracy of the world model.

[0088] As an optional implementation, before the world model issues the recommended strategy set to the global-level control agent based on the current traffic state and the traffic control causal graph of the target road network in step S102 above, the following steps are included: Construct a causal graph for traffic control.

[0089] The nodes in the traffic control causal graph include traffic scene feature nodes and global control strategy nodes, and the edges in the traffic control causal graph are used to represent the causal relationships between the nodes.

[0090] Optionally, a directed attribute causal graph can be constructed with the root causes of congestion, global control strategies, and implementation effects as links, serving as a causal graph for traffic control.

[0091] The traffic control causal graph includes four types of nodes: traffic scenario characteristic nodes, congestion root cause nodes, global control strategy nodes, and execution effect nodes. Specifically, traffic scenario characteristic nodes store attributes such as time period, weather, road network traffic, and disturbance type; congestion root cause nodes store attributes such as congestion type, impact range, and severity; global control strategy nodes store attributes such as timing schemes, coordination rules, and diversion schemes; and execution effect nodes store attributes such as traffic efficiency improvement rate, congestion dissipation time, and secondary congestion risk.

[0092] In the causal graph of traffic control, the edges are directed edges, which are used to represent the causal relationship between each node. The weight of the edge is the causal effect value. The strength of the causal relationship between each adjacent node is quantified by calculating the average treatment effect (ATE) between adjacent nodes and using it as the weight value.

[0093] It is worth noting that after each control task is completed, based on the characteristics of the traffic scenario, the root causes of congestion, the target global control strategy, and the execution effect, nodes and edges are added to the traffic control causal graph. By updating the causal weights of the edges, the control case is archived in the traffic control causal graph to provide reference data for subsequent traffic control.

[0094] In this embodiment, a traffic control causal graph is constructed. The nodes in the traffic control causal graph include traffic scenario feature nodes and global control strategy nodes. The edges in the traffic control causal graph are used to represent the causal relationships between the nodes. The traffic control causal graph facilitates the rapid transfer of historical traffic control cases, improving the efficiency and accuracy of strategy recommendation in the world model.

[0095] Figure 7 Flowchart of the traffic control method provided in the embodiments of this application Figure 6 ,like Figure 7 As shown, in step S201 above, the world model generates multiple candidate traffic control strategies based on the current traffic state and the causal graph of traffic control, including: S601. The world model extracts the causal feature vector of the current traffic scenario based on the current traffic state, and determines the similarity between the causal feature vector of the current traffic scenario and the feature vector of each historical traffic scenario feature node in the traffic control causal graph.

[0096] Optionally, the world model extracts feature vectors of the current traffic scenario based on the current traffic state, including 128-dimensional features such as traffic flow distribution, congestion type, and disturbance type. It then uses a Bidirectional Encoder Representations from Transformers (BERT) embedded model to generate causal feature vectors of the current traffic scenario.

[0097] In the causal graph of traffic control, the cosine similarity between the causal feature vector of the current traffic scenario and the feature vector of each historical traffic scenario feature node in the causal graph of traffic control is calculated.

[0098] S602. The world model generates multiple candidate traffic control strategies based on the similarity scores.

[0099] Optionally, the world model selects target global control strategies associated with multiple historical traffic scenario feature nodes with cosine similarity greater than the third threshold from the traffic control causal graph based on each cosine similarity, and generates multiple candidate traffic control strategies for the current traffic scenario based on the selected target global control strategies.

[0100] For example, the world model can use the target global control strategy associated with the top three historical traffic scenario feature nodes in the traffic control causal graph with a cosine similarity greater than or equal to 85% as a candidate traffic control strategy.

[0101] The world model can also adaptively optimize and adjust the target global control strategy based on the three historical traffic scenario feature nodes with a cosine similarity greater than or equal to 85%, generating three new sets of strategies as candidate traffic control strategies.

[0102] In this embodiment, the world model extracts the causal feature vector of the current traffic scenario based on the current traffic state, determines the similarity between the causal feature vector of the current traffic scenario and the feature vectors of feature nodes of each historical traffic scenario in the traffic control causal graph, and generates multiple candidate traffic control strategies based on the similarity scores. This enables rapid adaptation to various traffic scenarios and improves the efficiency and accuracy of traffic control.

[0103] Based on the same inventive concept, this application also provides a traffic control device corresponding to the traffic control method. Since the principle of the device in this application is similar to that of the traffic control method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0104] Figure 8 A modular structure diagram of the traffic control device provided in the embodiments of this application is shown below. Figure 8 As shown, the device includes: Module 801 is used to construct a hierarchical heterogeneous intelligent agent set corresponding to the target road network. The hierarchical heterogeneous intelligent agent set includes a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents. The construction module 801 is also used to construct a world model, which, based on the current traffic status and the traffic control causal graph of the target road network, distributes a set of recommended strategies to the global-level control agent. The generation module 802 is used to generate a global control strategy based on a recommended strategy set by a global-level management and control agent, and to solve the equilibrium of a master-slave game based on the global control strategy to obtain a target global control strategy. Multiple regional-level coordination agents and multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, and to perform traffic control based on each intersection control strategy.

[0105] As an optional implementation, the construction module 801 is specifically used for: A world model is trained based on rules from multiple traffic physics models, topological data of the target road network, historical traffic control data of the target road network, and traffic rule data.

[0106] As an optional implementation, the construction module 801 is specifically used for: Based on the current traffic conditions and the causal graph of traffic control, the world model generates multiple candidate traffic control strategies; The pre-built traffic control rule engine verifies each candidate traffic control strategy and determines at least one candidate traffic control strategy to be verified based on the verification results. The simulation results of each candidate traffic control strategy to be verified are determined, a recommended strategy set is generated based on the simulation results, and the recommended strategy set is distributed to the global-level management and control agent.

[0107] As an optional implementation, the generation module 802 is specifically used for: The global-level control agent evaluates the global traffic efficiency, secondary congestion risk, and stability of the disturbance scenario of each recommended strategy based on the simulation results of each recommended strategy in the recommended strategy set, and determines the comprehensive score of each recommended strategy. The global-level management agent selects the recommended strategy with the highest overall score as the global control strategy.

[0108] As an optional implementation, the generation module 802 is specifically used for: The global-level management agent solves the master-slave game equilibrium based on the global control strategy and the payoff functions of each layer of the hierarchical heterogeneous multi-agent, and obtains the target global control strategy. The target global control strategy is then distributed to each regional-level coordination agent. The target global control strategy includes global coordination rules, payoff function constraints, management priority, global objective, timing adjustment boundary, and timing constraints. Each regional coordination agent generates a regional coordination strategy based on the target global control strategy, the execution status of the local traffic strategy, and the revenue function of the execution agents at each intersection level under its jurisdiction, and then distributes the strategy to the execution agents at each intersection level under its jurisdiction. Each intersection-level execution agent generates a corresponding intersection control strategy based on the regional coordination strategy and the reward function of the intersection-level execution agent.

[0109] As an optional implementation, the construction module 801 is also used for: The differences between the simulation results at preset time intervals and the measured traffic conditions under the target global control strategy are statistically analyzed. If the difference information meets the preset conditions, traffic scene features are extracted, and the model parameters of the world model are adjusted according to the traffic scene features, the target global control strategy, and the measured traffic state under the target global control strategy to obtain the adjusted world model.

[0110] As an optional implementation, the construction module 801 is also used for: A causal graph for traffic control is constructed. The nodes in the causal graph include traffic scenario feature nodes and global control strategy nodes. The edges in the causal graph are used to represent the causal relationships between the nodes.

[0111] As an optional implementation, the construction module 801 is specifically used for: The world model extracts the causal feature vector of the current traffic scenario based on the current traffic state, and determines the similarity between the causal feature vector of the current traffic scenario and the feature vector of each historical traffic scenario feature node in the traffic control causal graph. The world model generates multiple candidate traffic control strategies based on their similarity scores.

[0112] This application also provides an electronic device, such as... Figure 9 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application, including a processor 91, a memory 92, and a bus 93. The memory 92 stores machine-readable instructions executable by the processor 91. When the electronic device is running, the processor 91 communicates with the memory 92 via the bus 93, and the processor 91 executes the machine-readable instructions to perform the steps of the traffic control method in the aforementioned embodiment.

[0113] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the traffic control method described in the foregoing embodiments.

[0114] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0115] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0116] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A traffic control method, characterized in that, include: Construct a hierarchical heterogeneous intelligent agent set corresponding to the target road network. The hierarchical heterogeneous intelligent agent set includes a global-level control intelligent agent, multiple regional-level coordination intelligent agents, and multiple intersection-level execution intelligent agents. A world model is constructed, and based on the current traffic state and the traffic control causal graph of the target road network, the world model issues a set of recommended strategies to the global-level control agent. The global-level management agent generates a global control strategy based on the recommended strategy set, and solves the master-slave game equilibrium according to the global control strategy to obtain the target global control strategy. The multiple regional-level coordination agents and the multiple intersection-level execution agents generate multiple intersection control strategies based on the target global control strategy, and perform traffic control according to each intersection control strategy.

2. The method according to claim 1, characterized in that, The construction of the world model includes: The world model is trained based on the rules of multiple traffic physics models, the topology data of the target road network, the historical traffic control data of the target road network, and traffic rule data.

3. The method according to claim 1, characterized in that, The world model, based on the current traffic state and the traffic control causal graph of the target road network, issues a set of recommended strategies to the global-level control agent, including: Based on the current traffic state and the traffic control causal graph, the world model generates multiple candidate traffic control strategies. The pre-built traffic control rule engine verifies each candidate traffic control strategy and determines at least one candidate traffic control strategy to be verified based on the verification results. The simulation results of each candidate traffic control strategy to be verified are determined, the recommended strategy set is generated based on the simulation results, and the recommended strategy set is distributed to the global-level management and control agent.

4. The method according to claim 1, characterized in that, The generation of a global control policy by the global-level management agent based on the recommended policy set includes: The global-level management and control agent evaluates the global traffic efficiency, secondary congestion risk, and stability of the disturbance scenario of each recommended strategy based on the simulation results of each recommended strategy in the recommended strategy set, and determines the comprehensive score of each recommended strategy. The global-level management agent selects the recommended strategy with the highest comprehensive score as the global control strategy.

5. The method according to claim 1, characterized in that, The process involves solving a master-slave game equilibrium based on the global control strategy to obtain a target global control strategy. Then, the multiple regional-level coordinating agents and the multiple intersection-level executing agents generate multiple intersection control strategies based on the target global control strategy, including: The global-level management agent performs a master-slave game equilibrium solution based on the global control strategy and the benefit functions of each layer of the hierarchical heterogeneous multi-agent agent to obtain the target global control strategy, and then distributes the target global control strategy to each regional-level coordination agent. The target global control strategy includes global coordination rules, benefit function constraints, management priority, global objective, timing adjustment boundary and timing constraints. Each regional coordination agent generates a regional coordination strategy based on the target global control strategy, the execution status of the local traffic strategy, and the revenue function of the execution agents at each intersection level under its jurisdiction, and then distributes the strategy to the execution agents at each intersection level under its jurisdiction. Each intersection-level execution agent generates a corresponding intersection control strategy based on the regional coordination strategy and the reward function of the intersection-level execution agent.

6. The method according to claim 1, characterized in that, The method further includes: The differences between the simulation results at preset time intervals and the measured traffic conditions under the target global control strategy are statistically analyzed. If the difference information meets the preset conditions, traffic scene features are extracted, and the model parameters of the world model are adjusted according to the traffic scene features, the target global control strategy, and the measured traffic state under the target global control strategy to obtain the adjusted world model.

7. The method according to claim 1, characterized in that, Before the world model issues the recommended strategy set to the global-level management agent based on the current traffic state and the traffic control causal graph of the target road network, the process includes: The traffic control causal graph is constructed, and the nodes in the traffic control causal graph include traffic scene feature nodes and global control strategy nodes. The edges in the traffic control causal graph are used to represent the causal relationships between the nodes.

8. The method according to claim 3, characterized in that, The world model generates multiple candidate traffic control strategies based on the current traffic state and the traffic control causal graph, including: The world model extracts the causal feature vector of the current traffic scenario based on the current traffic state, and determines the similarity between the causal feature vector of the current traffic scenario and the feature vector of each historical traffic scenario feature node in the traffic control causal graph. The world model generates multiple candidate traffic control strategies based on their similarity scores.

9. An electronic device, characterized in that, include: The electronic device includes a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the traffic control method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the traffic control method as described in any one of claims 1 to 8.

Images