Main line traffic flow management and control optimization method based on high-performance traffic simulation agent

Abstract

A main line traffic flow management and control optimization method based on a high-performance traffic simulation agent comprises the steps that a target area road network is selected, and a main line road section and intercommunication road section structure is constructed; a multi-scene sample set is generated through microscopic traffic simulation, and traffic operation states under different traffic requirements, speed limiting and hard shoulder open conditions are simulated; training a neural network model based on a deep learning method, and constructing an efficient traffic simulation agent by taking the target road section and upstream and downstream multilayer associated topological information as input features and taking travel time and average speed as labels; and in combination with a multi-objective optimization algorithm, taking total travel time and carbon emission as optimization objectives, and dynamically iterating to generate an optimal management and control scheme of road section speed limitation and hard road shoulder opening. According to the method, traditional low-efficiency microscopic simulation is replaced with the simulation agent, the calculation performance is remarkably improved, and collaborative optimization of the complex road network and scientific formulation of a low-carbon management and control strategy are achieved.

Description

Technical Field

[0001] The present invention relates to traffic control and artificial intelligence technology, and in particular to a mainline traffic flow control optimization method based on a high-performance traffic simulation intelligent agent. Background Art

[0002] Traffic control plays a vital role in balancing traffic network loads, alleviating congestion, improving transportation efficiency, and reducing accidents. On highways, traffic control typically includes measures such as speed limits, opening hard shoulders, and traffic guidance. How to scientifically and rationally formulate traffic control plans is a key issue in traffic management. Currently, methods for formulating highway traffic control plans typically include those based on manual experience, field-measured road data, traffic simulation models, optimization algorithms, and reinforcement learning. Manual experience relies on long-term observation of the road network by relevant traffic management personnel, resulting in relatively rigid control plans. These plans lack flexibility and portability, and the inherent complexity of the traffic system makes empirical analysis difficult. Intelligent optimization methods based on field-measured road data, particularly reinforcement learning, require the system to possess a certain degree of trial-and-error learning capability. However, safety constraints in traffic systems dictate avoiding high-risk trial-and-error, making their practical implementation limited. Intelligent optimization methods based on traffic simulation offer theoretically greater adaptability. After calibrating the traffic simulation system, they can achieve a full-sample estimation of the traffic operating conditions on the real road network. These methods can then be combined with various optimization algorithms to generate optimal control strategies. In terms of traffic simulation, microscopic traffic simulation offers greater precision and interpretability, helping to accurately simulate the impact of various control measures. However, its operational efficiency is low, making it difficult to support computationally intensive intelligent optimization algorithms. This has led to its limited application in the intelligent generation of traffic control strategies. While there are currently technologies to improve the performance of microscopic traffic simulation, these technologies primarily rely on hardware enhancements and the introduction of parallel computing.

[0003] Specifically, the following technical problems exist in the formulation of existing highway traffic flow control plans: 1. Traditional highway traffic flow control schemes rely heavily on manual experience: Due to the large scale, multiple influencing factors, and complex spatiotemporal correlations inherent in highway traffic flow control, relying solely on manual experience to formulate plans is difficult, requires long-term observation, and requires high professional and technical skills from managers. Furthermore, the resulting plans and rules are difficult to quickly transfer to other regional road networks.

[0004] 2. Optimization methods based on measured data and reinforcement learning have limited application in transportation systems: Reinforcement learning is a classic and effective method in the field of optimization and control, but it generally requires the system to have a certain trial-and-error learning cost. Safety constraints in transportation systems require avoiding high-risk trial and error, so its practical implementation is not feasible.

[0005] 3. Difficulties in applying micro-traffic simulation in traffic flow control and optimization: Micro-traffic simulation offers significant advantages in being more precise, intuitive, and explanatory when simulating the impact of various control measures on traffic system operations and estimating system states. However, due to its low operational efficiency, it is difficult to support various optimization algorithms with high computational requirements, making its application in traffic flow control and optimization more difficult.

[0006] To address the issues of over-reliance on human experience in developing highway traffic flow control plans, the limited feasibility of reinforcement learning methods, and the low efficiency and application difficulties of micro-traffic simulation, existing technical solutions include: Macroscopic traffic flow models are used as a means of estimating traffic system states, using speed, flow, and density as core parameters to predict road segment traffic conditions. However, due to the lack of comprehensive and sufficient measured traffic data, some studies still calibrate macroscopic traffic flow models based on microscopic traffic simulation results. Furthermore, most macroscopic traffic flow models are not good at reflecting the mutual influence of traffic flows between sections within the road network, and they inadequately model the systematic and coordinated laws of the traffic system, which can easily lead to distortion.

[0007] With the help of parallel computing technology, the running tasks of micro-traffic simulation are decomposed to achieve parallel simulation, so as to improve the running speed of micro-traffic simulation. However, this method places higher requirements on the hardware facilities of the running model, and the effect of improving the running speed is limited. It is still difficult to meet the various optimization algorithms with high computing requirements.

[0008] It should be noted that the information disclosed in the above background technology section is only used to understand the background of this application, and therefore may include information that does not constitute prior art known to ordinary technicians in this field. Summary of the Invention

[0009] The main purpose of the present invention is to overcome the defects existing in the above-mentioned background technology and provide a main line traffic flow control optimization method based on high-performance traffic simulation intelligent agent.

[0010] To achieve the above object, the present invention adopts the following technical solutions: A mainline traffic flow control optimization method based on a high-performance traffic simulation agent includes the following steps: S1. Select the target study area and determine the regional road network consisting of main line sections and interchange sections; S2. Multi-scenario sample generation based on microscopic traffic simulation: For the target study area, multiple traffic scenarios with different traffic demand levels, road speed limits, and hard shoulder availability are set. Traffic operation status indicators for road sections under each scenario are generated through batch simulation to form a traffic simulation sample set; S3. Build a high-performance traffic simulation agent: Based on the traffic simulation sample set, a deep learning method is used to train a neural network model. The model uses the upstream and downstream topological information and its own attributes of the target road section as input features, and the traffic operation status indicators of the road section as labels to achieve efficient evaluation of the road network traffic status; S4. Optimize the control plan by combining a multi-objective optimization algorithm: take the total travel time and total carbon emissions of the main line as optimization objectives, and the speed limit and hard shoulder opening status of each road section as decision variables, use the traffic simulation agent to evaluate candidate plans, and iterate the optimization algorithm to generate the optimal control plan.

[0011] A computer program product includes a computer program, which implements the main line traffic flow control optimization method when executed by a processor.

[0012] The present invention has the following beneficial effects: The present invention proposes a mainline traffic flow control optimization method based on a high-performance traffic simulation agent. By combining micro-traffic simulation with deep learning technology, a high-performance traffic simulation agent is constructed, which effectively solves the problems of strong dependence on traditional manual experience, high trial-and-error risk of reinforcement learning, and low efficiency of micro-simulation. The method of the present invention uses multi-scenario batch simulation to generate massive samples, extracts spatial topological information and attribute characteristics of the target road section and its upstream and downstream multi-layer associated road sections through recursive traversal, and combines neural network (such as multi-layer perceptron (MLP)) training to achieve efficient evaluation of traffic status, significantly improving simulation speed without relying on hardware upgrades. It further integrates multi-objective optimization algorithms, takes travel time and carbon emissions as dual optimization objectives, and dynamically generates speed limit control and hard shoulder opening plans, taking into account the operating efficiency of the traffic system and low-carbon needs, and verifies the reliability of the plan through micro-simulation replication, ultimately forming a scientific and transferable control strategy, breaking through the limitations of traditional methods in collaborative optimization and real-time response of complex road networks.

[0013] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is the overall flow chart of the mainline traffic flow control optimization method based on high-performance traffic simulation intelligent agent of the present invention.

[0015] Figure 2 A flow chart for generating multi-scenario simulation samples according to an embodiment of the present invention.

[0016] Figure 3 This is a flowchart of the multi-objective genetic algorithm (NSGA-II) optimization of an embodiment of the present invention.

[0017] Figure 4 Schematic diagram of the road network in the target study area according to an embodiment of the present invention.

[0018] Figure 5 This is a comparison chart of the implementation effects of the variable speed limit control solution in an embodiment of the present invention.

[0019] Figure 6 This is a comparison chart of the implementation effects of the hard shoulder opening control solution according to an embodiment of the present invention.

[0020] Figure 7 This is a comparison chart of the optimization effects of the integrated management and control measures in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope of the present invention and its application.

[0022] The present invention aims to solve the technical problems of low operating efficiency of micro-traffic simulation and difficulty in quickly estimating the state of the traffic system in the intelligent optimization of traffic control strategies. It provides a main line traffic flow control optimization method based on a high-performance traffic simulation intelligent agent. By training and learning a high-performance traffic simulation intelligent agent based on a large sample set generated by micro-traffic batch simulation, it fully combines the advantages of micro-traffic simulation and solves its low operating efficiency problem. It is further combined with a multi-objective optimization algorithm (such as a genetic optimization algorithm) to generate specific plans for control measures such as variable speed limits on high-speed main lines and open hard shoulders, thereby assisting actual high-speed traffic control decisions and improving the efficiency of the traffic system.

[0023] See Figure 1 The embodiment of the present invention provides a main line traffic flow control optimization method based on a high-performance traffic simulation agent, comprising the following steps: Step S1: Select a target research area and determine a regional road network consisting of main line segments and interchange segments.

[0024] Step S2, generate multi-scenario samples based on micro-traffic simulation: for the target study area, set up multiple traffic scenarios with different traffic demand levels, road speed limits, and hard shoulder opening conditions, and generate traffic operation status indicators for road sections under each scenario through batch simulation to form a traffic simulation sample set.

[0025] In some embodiments, in step S2, generating multi-scene samples includes: Based on three indicators: route travel volume, road speed limit and hard shoulder open status, different traffic scenarios are randomly set within a preset range. Each scenario is divided into two stages, and the indicators of the two stages are independently and randomly adjusted to simulate the impact of dynamic changes in traffic demand and control measures.

[0026] Step S3, constructing a high-performance traffic simulation intelligent agent: Based on the traffic simulation sample set, a deep learning method is used to train a neural network model. The model uses the upstream and downstream topological information and its own attributes of the target road section as input features, and the traffic operation status indicators of the road section as labels to achieve efficient evaluation of the road network traffic status.

[0027] In some embodiments, in step S3, the method for extracting upstream and downstream topology information includes: Perform multi-layer recursive traversal on the target road section interval, obtain the attributes of its upstream and downstream associated road sections layer by layer, aggregate the features of the associated road sections at each layer, and form a feature vector that represents the spatial dependency relationship.

[0028] In some embodiments, the specific process of the multi-layer recursive traversal includes: a) First-level search: For the target road section, obtain the set of sections of its directly adjacent upstream entry, upstream exit, downstream entry, and downstream exit sections; b) Second-level search: For each road segment interval in the first-level set, recursively search its directly adjacent upstream and downstream road segment intervals to generate a secondary road segment interval set; c) Third-level search: For each road segment interval in the second-level set, continue to recursively search its directly adjacent upstream and downstream road segment intervals to generate a third-level road segment interval set; d) Feature aggregation: The attribute features of the road sections in each level set are aggregated by level to form a multidimensional feature vector that represents the spatial dependency relationship.

[0029] In some embodiments, the feature aggregation specifically includes: for each level of the road segment interval set generated by recursive traversal, respectively calculating the feature mean of the traffic demand, inherent attributes of the road segment and control parameters of the road segment interval in the set; Among them, the inherent attributes of the road segment include the number of lanes, length and historical traffic volume; Control parameters include speed limit and hard shoulder open status; The mean results of the sets of each level are flattened and concatenated in hierarchical order to form a multi-dimensional feature vector as the input feature of the neural network model.

[0030] In some embodiments, the neural network model is a multi-layer perceptron (MLP), which generates a traffic simulation agent by learning the mapping relationship between the feature vector and the traffic operation status indicator through training.

[0031] Step S4: Optimizing the control plan by combining a multi-objective optimization algorithm: taking the total travel time of the main line and the total carbon emissions as optimization objectives, and the speed limit and hard shoulder opening status of each road section as decision variables, using the traffic simulation agent to evaluate candidate plans, and iterating the optimization algorithm to generate the optimal control plan; In some embodiments, in step S4, the multi-objective optimization algorithm adopts a genetic algorithm of the NSGA-II framework, iteratively optimizes the population through non-dominated sorting, selection, crossover and mutation operations, and quickly evaluates the fitness of individuals using the traffic simulation agent.

[0032] In some embodiments, the method for calculating the total travel time of the main line is: based on the accumulation of the product of traffic demand and travel time of each road section; the method for calculating the total carbon emissions is: based on the product of traffic demand, carbon emission factor and road section length of each road section.

[0033] In some embodiments, in the calculation of the total travel time, the traffic demand of the road section is obtained by weighted summation of the path travel volume and the path-road section mapping relationship; the calculation of the total carbon emissions specifically includes: calculating the carbon emissions of each vehicle model based on the traffic demand, basic energy consumption, speed correction coefficient, road section length and unit energy consumption emission factor of different vehicle models within the road section, and accumulating the results of all vehicle models.

[0034] Furthermore, the mainline traffic flow control optimization method based on high-performance traffic simulation agent also includes: Step S5: Verify the control plan: Perform a microscopic traffic simulation on the control plan generated by the multi-objective optimization algorithm to verify its effectiveness. The control plan output by the multi-objective optimization algorithm is input into the microscopic traffic simulation model and the simulation results are compared with the agent's prediction results to verify the effectiveness of the plan.

[0035] The core of this invention is to find a method to integrate the advantages of micro-traffic simulation in traffic flow control optimization while avoiding the problem of poor performance of micro-traffic simulation. A traffic simulation intelligent agent is trained based on a large-scale micro-traffic simulation sample set. The intelligent agent has extremely high prediction performance for road network traffic status. At the same time, its prediction results are highly consistent with the output of micro-traffic simulation, and can reproduce the control scenarios through micro-traffic simulation.

[0036] The specific innovations of the present invention include: 1. Construction of a large-scale micro-traffic simulation sample set: For the target study area, batch simulations of traffic scenarios with varying traffic demands, speed limits, and hard shoulder availability are conducted. Each traffic scenario is simulated in two phases, with three key indicators used in each phase taking independent random values. This allows for the simulation of traffic state transitions following changes in traffic demand and control measures.

[0037] 2. Construction of a high-performance traffic simulation agent: Taking road sections as prediction targets, a recursive traversal method is used to obtain the multi-layer spatial correlation between upstream and downstream sections of the target road section. Sample features are designed based on the spatial correlation between upstream and downstream sections and the characteristic attributes of the sections themselves. The designed sample features and sample labels are input into the neural network for training and learning, thereby constructing a high-performance traffic simulation agent.

[0038] 3. Mainline traffic flow control optimization method based on high-performance traffic simulation agent: The total travel time and total carbon emissions of the mainline are used as optimization objectives, the speed limit of each section of the mainline and whether the hard shoulder is open are used as decision variables, and a high-performance traffic simulation agent is used as a feedback model for the control plan effect. Combined with a genetic algorithm, a method for optimizing the mainline traffic flow control plan is constructed.

[0039] In the specific implementation process, the present invention first selects the target research area: that is, selects the regional road network composed of the main line section and the interchange road sections it passes through.

[0040] Then, multi-scenario samples are generated based on micro-traffic simulation. That is, micro-traffic simulation is carried out for various scenarios such as different traffic demand levels, different road speed limits, and different hard shoulder opening conditions in the target study area. The traffic operation status (travel time, average speed, etc.) of each road section in the study area under each scenario is obtained to form a relatively complete traffic simulation sample set.

[0041] Based on a large-scale microscopic traffic simulation sample set, a data-driven approach and deep learning methods are used to train and learn the data, thereby constructing a traffic simulation agent that can accurately and effectively assess the traffic operation status of the road network under different scenarios. Before training the samples, a recursive traversal method is used to obtain the multi-layer spatial correlation between upstream and downstream sections of the target sample section. Sample features are designed based on the spatial correlation between upstream and downstream sections and the characteristic attributes of the sections themselves.

[0042] Finally, a multi-objective genetic algorithm is used to optimize the control scheme. That is, based on the high-performance traffic simulation intelligent agent that has been constructed, a multi-objective genetic algorithm is used to intelligently generate speed limit and hard shoulder opening schemes for each section of the main line. The control effect is then re-verified by performing corresponding micro-traffic simulation on the control scheme.

[0043] The present invention has significant advantages in the intelligent optimization of traffic flow control schemes. First, its estimation of the state of the traffic system is still essentially based on micro-traffic simulation, making full use of the advantages of micro-traffic simulation being more sophisticated, intuitive, and highly interpretable. Secondly, by training and learning on the micro-traffic simulation sample set, a traffic simulation intelligent agent is constructed. Its prediction results of the traffic system state are highly consistent with those of the micro-traffic simulation, but the computing performance is greatly increased. This significant performance improvement enables it to be combined with a genetic optimization algorithm, ultimately achieving automatic optimization of the traffic flow control scheme. The optimization results of the control scheme can be easily reproduced in a scene through micro-traffic simulation.

[0044] The present invention achieves these advantages primarily due to the successful training of a highly accurate micro-traffic simulation agent on a large-scale micro-traffic simulation sample set. The inputs to micro-traffic simulation are typically the road network, the attributes of each road section (speed limit, availability of hard shoulders, etc.), travel routes, travel demand, etc., and the outputs are traffic status indicators (average speed, flow rate, etc.) for each road section in the road network. Successfully learning this input-output mapping through deep learning is very difficult due to the complexity of the inputs, particularly inputs such as the road network and travel routes, which are graph data containing rich spatial correlations. The present invention conducts in-depth research on the spatial correlations between each road section in the road network and the spatiotemporal patterns of traffic flow. Based on the spatial correlations between upstream and downstream sections and the characteristic attributes of the sections themselves, a method for extracting sample features was designed. This allows the successful construction of a traffic simulation agent through a data-driven, deep learning approach. Furthermore, combined with a genetic optimization algorithm, automatic optimization of traffic flow control schemes is achieved.

[0045] The following further describes specific embodiments of the present invention, its algorithm examples and experimental verification.

[0046] To address the technical issues of low operational efficiency in micro-traffic simulations and the difficulty in rapidly estimating traffic system states in intelligent optimization of traffic control strategies, the present invention provides the following technical solution: Based on a large sample set generated by batch micro-traffic simulations, a high-performance traffic simulation agent is trained and learned to fully leverage the advantages of micro-traffic simulation while avoiding its low operational efficiency. This is further combined with a genetic optimization algorithm to generate specific solutions for control measures such as variable speed limits on expressways and open hard shoulders. The mainline traffic flow control optimization method based on a high-performance traffic simulation agent includes the following steps: S1. Select the target research area: that is, select the regional road network consisting of the main line section and the interchange sections it passes through; S2. Multi-scenario sample generation based on micro-traffic simulation: Conduct micro-traffic simulations for various scenarios, including varying traffic demand levels, speed limits, and hard shoulder availability, for the target study area. The traffic status (travel time, average speed, etc.) for each road section within the study area is obtained under each scenario, forming a relatively complete traffic simulation sample set. S3. Build a high-performance traffic simulation agent based on traffic simulation sample sets: Using a data-driven approach, deep learning methods are used to train and learn from a large number of traffic simulation samples to build a traffic simulation agent that can accurately and efficiently evaluate the traffic operation status of the road network in different scenarios. S4. Optimize control plans by combining a multi-objective genetic algorithm: Based on the established high-performance traffic simulation agent, a multi-objective genetic algorithm is used to intelligently generate speed limit and hard shoulder opening plans for each section of the main line. S5. Re-verify the control scheme output by the genetic algorithm: After the genetic algorithm quickly optimizes and generates multiple control schemes, the control effect is re-verified by performing corresponding micro-traffic simulation on the control schemes.

[0047] Furthermore, in step S2, different traffic scenarios are set for simulation based on three key indicators: path travel volume, section speed limit, and whether the section has a hard shoulder. Among them, the path travel volume is randomly selected in the range of 0~1000 pcu / h, the section speed limit is randomly selected in the range of 30~120 km / h, and the section hard shoulder is randomly selected in terms of whether it is open. Each simulation scenario is simulated for 1 hour and is divided into two stages: the first 0.5h and the last 0.5h. The three key indicators taken in each stage are independently randomly selected, so as to realize the simulation of traffic state switching after changes in traffic demand and control measures during the simulation process. For the basic process of generating multi-scenario samples based on micro-traffic simulation, please refer to Figure 2 .

[0048] Furthermore, in step S3, the traffic simulation agent models each road segment. Its sample labels (directly predicted targets) are the average speed and travel time for the segment. Sample features comprehensively consider the upstream and downstream topological information of the segment, as well as the segment's own characteristic attributes. Regarding the upstream and downstream topological information of the segment, the agent recursively traverses the upstream and downstream related segments of the target segment to capture the spatial dependencies between traffic operations across different segments in the road network. For the segment's own characteristic attributes, six features are considered: the number of lanes, length, traffic flow in the first phase (representing the segment's historical traffic status), traffic demand in the second phase, whether the hard shoulder is open, and speed limit (representing the segment's future traffic demand and implemented control measures).

[0049] Furthermore, the algorithm for traversing the upstream and downstream related road sections of the target road section is specifically described as follows: S3-1. For the target road section, search for the upstream entry road section, upstream exit road section, downstream entry road section, and downstream exit road section adjacent to the target road section, and group them into four sets respectively. This search process is the first-level search.

[0050] S3-2. For the upstream entry segment interval set obtained from the first-level search, search for the upstream entry segment interval and upstream exit segment interval of each segment interval in the set, and group them into two sets respectively. For the upstream exit segment interval set obtained from the first-level search, search for the downstream entry segment interval and downstream exit segment interval of each segment interval in the set, and group them into two sets respectively. For the downstream entry segment interval set obtained from the first-level search, search for the upstream entry segment interval and upstream exit segment interval of each segment interval in the set, and group them into two sets respectively. For the downstream exit segment interval set obtained from the first-level search, search for the downstream entry segment interval and downstream exit segment interval of each segment interval in the set, and group them into two sets respectively. This search process generates a total of 8 new segment interval sets, which constitute the second-level search.

[0051] S3-3. For the 8 road segment interval sets obtained by the second-level search, use the same search method as S3-2 to search the relevant upstream and downstream road segment intervals of each road segment interval in each set, and generate 16 new road segment interval sets. This process is the third-level search.

[0052] Furthermore, after the algorithm is executed to traverse the upstream and downstream related sections, the final output is recorded as , including a total of 29 road section interval sets (The target road section is one of the sets). The number of road sections in each set varies according to the topology of the road network. The characteristics of the road sections in each set are summed and averaged, and then the whole set is averaged. Flatten. Finally, the feature of each sample is a A character vector of length .

[0053]

[0054]

[0055] Where, It is a set of 29 road sections output by the upstream and downstream related road section traversal algorithm; for A set of road segment intervals contained in ; For collection The number of road sections included in ; is a collection The average traffic demand of the road segments included in the set; is a collection Average open hard shoulder markings (average of the open hard shoulder markings for the road sections included in the set); is a collection The average speed limit (the average of the speed limits of the road sections included in the set); is a collection The average number of lanes (the average number of lanes in the road sections included in the set); is a collection The average length of the road segment (the average length of the road segment intervals contained in the set); is a collection The average traffic flow of the previous adjacent period (the average traffic flow of the previous adjacent period of the road sections included in the set).

[0056] Furthermore, after constructing the traffic simulation sample set, an MLP (Multi-Layer Perceptron) neural network is built to train and learn the mapping relationship between the features and labels of each sample, thus forming a high-performance traffic simulation intelligent agent.

[0057] Furthermore, in step S4, the mainline traffic optimization objectives include total travel time and total carbon emissions.

[0058] The total travel time for the main line is calculated as follows:

[0059]

[0060] Where, Refers to the total travel time of the main line; Main line section Total travel time; Main line section travel time; Main line section Hourly traffic demand; The paths involved in the road network Hourly travel volume; It is the set of all paths involved in the road network; Indicates the main line section With path The corresponding relationship, if the path Passing through the main line section 1 if yes, 0 otherwise.

[0061] The calculation method for the main line's total carbon emissions is as follows:

[0062]

[0063] Where, Refers to the total carbon emissions of the main line; Main line section Total carbon emissions; Main line section Carrying vehicle models Hourly transportation demand; Main line section mid-size car The carbon emissions are calculated as follows.

[0064]

[0065] Where, Refers to the model In the road section Carbon emissions generated by operation (kg); For car models Basic energy consumption, the unit for fuel vehicles is L / km, and the unit for electric vehicles is kwh / km; For car models In the road section The average speed is Correction factor when For road section Length (km); For car models The unit of carbon dioxide emission factor for fuel vehicles is kg / L, and the unit for electric vehicles is kg / kwh.

[0066] Furthermore, in step S4, the classic multi-objective genetic algorithm NSGA-II is used, and each individual represents a control plan, which is specifically composed of attributes such as the speed limit value of each main line section or whether the hard shoulder is open; the adaptability of each individual to the environment is measured by two indicators: the total travel time of the main line and the total carbon emissions. In each round of iteration, the genetic algorithm performs non-dominated sorting, selection, crossover and mutation operations on multiple individuals according to individual fitness, and gradually obtains a better set of individuals. Under the iterative tuning of the NSGA-II algorithm, a better control plan is output, which is specifically manifested in the speed limit value that should be set for each section of the main line and whether the hard shoulder is open. For the process of the multi-objective genetic algorithm NSGA-II, see Figure 3 The specific steps are as follows: Initialize the population: generate an initial set of candidate solutions, each solution represents a control scheme; Iterative evolution: a) Genetic operation: perform selection, crossover and mutation operations on the current population to generate a child population; b) Merge population: merge the parent and child populations to form an extended population; c) Non-dominated sorting: perform fast non-dominated sorting on the merged population to divide the dominance level of individuals; d) Crowding calculation: calculate the crowding distance of individuals in the same non-dominated layer to ensure diversity; e) Elite selection: screen individuals according to non-dominated level and crowding degree to generate a new generation of parent population; Termination judgment: if the number of iterations does not reach the set threshold, repeat step 2; otherwise, output the optimal solution set.

[0067] Examples and experimental verification: In this example, a specific scheme of variable speed limit and open hard shoulder is intelligently optimized for a certain expressway main line section. The simulated road network of the selected area is as follows: Figure 4 As shown in the figure, the road network is divided into 114 directional sections based on the interchange nodes; the main line section is divided into 24 directional sections, and each section can be individually set with a speed limit and whether to open the hard shoulder.

[0068] The effectiveness evaluation of different control schemes requires a baseline scenario as a comparison. The baseline scenario defined in this chapter is: for the target study area, each route has a travel volume of 200 pcu / h, a mainline speed limit of 100 km / h, a ramp speed limit of 60 km / h, and no hard shoulder is available on any mainline section. In this example, the variable speed limits for each mainline section are 60 km / h, 70 km / h, 80 km / h, 90 km / h, 100 km / h, 110 km / h, and 120 km / h. The hard shoulder operating conditions for each mainline section are either open or closed.

[0069] For the selected road network scenario, we optimized and solved the variable speed limit control measures and analyzed the implementation effect of the plans. From the 10 batches of control plans generated by the genetic algorithm through multiple rounds of iterative optimization, we selected the plan with the highest carbon emissions in each batch and performed micro-traffic simulation on each plan to verify the prediction results. The results of the intelligent optimization algorithm are prediction results, while those based on micro-traffic simulation are simulation results. The implementation effects of the two methods can be seen in Figure 5 . After micro-traffic simulation verification, the intelligent optimization method of the control scheme has effectively reduced the total carbon emissions and total travel time of the main line through multiple rounds of iterative optimization. When the variable speed limit measure is implemented alone and the total carbon emissions indicator is dominant, the 8th scheme is better. Compared with the baseline scenario, it can reduce the total carbon emissions by 7.33% and the total travel time by 7.69%; compared with the scheme in which the speed limit of all sections of the main line is set to a maximum of 120km / h, it can reduce the total carbon emissions by 9.24% and the total travel time by 6.96%. The specific content of the 8th scheme is shown in Table 1.

[0070] Table 1

[0071] (1) Scenario analysis of hard shoulder opening For the selected road network scenario, we optimized and solved the hard shoulder opening control measures and analyzed the implementation effects. From the large number of solutions generated by the genetic algorithm through multiple rounds of iterative optimization, we selected 10 optimization solutions dominated by carbon emissions in the order of iteration. We then performed micro-traffic simulation on each solution to verify the prediction results. The results of the intelligent optimization algorithm are prediction results, while those based on micro-traffic simulation are simulation results. The implementation effects of the two methods can be found in Figure 6 . After micro-traffic simulation verification, the intelligent optimization method of the control scheme output the hard shoulder opening scheme in multiple rounds of iterative optimization, which effectively reduced the total carbon emissions and total travel time of the main line traffic. When the hard shoulder opening measure is implemented alone and the total carbon emissions of the main line traffic are dominant, the 10th scheme is better, which can reduce the total carbon emissions of the main line traffic by 9.84% and the total travel time of the main line by 22.30%; compared with the scheme in which all sections of the main line have hard shoulders, it can reduce the total carbon emissions of the main line traffic by 4.31% and the total travel time of the main line by 11.69%. The specific content of the 10th scheme is shown in Table 2.

[0072] Table 2

[0073] (2) Scenario analysis of integrated control measures For the selected road network scenario, the optimization solution of the fusion scheme of two types of control measures, variable speed limit and hard shoulder opening, and the implementation effect of the scheme are analyzed. Among the large number of schemes generated by the genetic algorithm through multiple rounds of iterative optimization, 10 optimization schemes dominated by carbon emissions are selected in the order of iteration, and micro-traffic simulation is performed on each scheme to verify the prediction results. The optimization effect of the fusion control measures is as follows: Figure 7 As shown in the figure, after micro-traffic simulation verification, the intelligent optimization method for control schemes, through multiple rounds of iterative optimization, output a fusion of control measures that effectively reduced total carbon emissions and total travel time for mainline traffic. When the fusion control scheme is based on the mainline traffic total carbon emissions indicator, the 10th scheme is superior, reducing total carbon emissions for mainline traffic by 13.00% and total travel time by 25.02%. See Table 3 for the details of the 10th scheme.

[0074] Table 3

[0075] Alternative embodiment: Based on the traffic simulation agent, the genetic algorithm can be replaced by other optimization algorithms, such as particle swarm optimization, ant colony optimization, simulated annealing algorithm, etc. It is even possible to dynamically generate traffic flow control plans by combining the traffic simulation agent with reinforcement learning.

[0076] In the training of traffic simulation intelligent agents, based on the sample feature extraction method proposed in the present invention based on the spatial correlation relationship between upstream and downstream road sections and the characteristic attributes of the road sections themselves, in addition to using conventional MLP neural networks for training, it can also be replaced by other neural network structures, such as attention networks.

[0077] The present invention is suitable for traffic flow control on expressway networks, but the control optimization method can also be applied to urban roads. As long as factors such as signal control are introduced into the sample characteristics, it is possible to predict the traffic operation status of urban roads.

[0078] This paper proposes a mainline traffic flow control optimization method based on a high-performance traffic simulation agent. Its innovative contributions and important features include: 1. Construction of a large-scale micro-traffic simulation sample set: For the target study area, batch simulations of traffic scenarios with different traffic demands, speed limits on different road sections, and different hard shoulder opening conditions are carried out. The simulation of each traffic scenario is divided into two stages, and the three key indicators adopted in each stage are independently randomly valued. In this way, the traffic state switching after changes in traffic demand and control measures can be simulated during the simulation process.

[0079] 2. Construction of a high-performance traffic simulation agent: Taking the road section as the prediction object, a recursive traversal method is used to obtain the multi-layer spatial correlation relationship between upstream and downstream sections of the target section. Sample features are designed based on the spatial correlation relationship between upstream and downstream sections and the characteristic attributes of the section itself. The designed sample features and sample labels are input into the neural network for training and learning, thereby constructing a high-performance traffic simulation agent.

[0080] 3. Mainline traffic flow control optimization method based on high-performance traffic simulation intelligent agent: The total travel time and total carbon emissions of the main line are used as optimization targets, the speed limit of each section of the main line and whether the hard shoulder is open are used as decision variables, and a high-performance traffic simulation intelligent agent is used as a model to feedback the effect of the control plan. Combined with the genetic algorithm, a method for optimizing the mainline traffic flow control plan is constructed.

[0081] The present invention has achieved remarkable technical effects in the intelligent generation and optimization of highway traffic control schemes. Compared with the background technology, it has the following advantages: in the intelligent optimization of traffic control schemes, the present invention fully utilizes the ability of micro-traffic simulation to perform refined simulations, but at the same time significantly improves the performance of traffic system state estimation by constructing a traffic simulation intelligent body, and does not rely on a large amount of hardware stacking, thereby making the application of micro-traffic simulation in the field of traffic control scheme optimization sufficiently feasible, and helping to achieve the scientific formulation of traffic control plans for the road network in the target area of control.

[0082] The realization of these technical effects is due to the multiple innovations of the present invention: First, through micro-traffic batch simulation, full-sample estimation of the traffic operation status of the target road network is achieved, and a large-scale traffic simulation sample set is constructed; second, each section in the road network is used as the prediction object, and the sample characteristics are constructed by the spatial correlation relationship between upstream and downstream sections and the characteristic attributes of the section itself. The total travel time and total carbon emissions of the section are used as sample labels, and the training of the neural network based on the large-scale traffic simulation sample set is successfully realized, thereby constructing a traffic simulation intelligent body, which has micro-traffic simulation capabilities while achieving extremely high computing performance; third, on the basis of the traffic simulation intelligent body, combined with the multi-objective optimization genetic algorithm, the scheme optimization of control measures such as main line variable speed limit and open hard shoulder is realized, thereby improving the main line traffic operation efficiency and reducing traffic carbon emissions.

[0083] Based on the above innovations, the present invention achieves multiple technical effects of high performance, improved traffic efficiency, and carbon reduction through micro-traffic batch simulation, intelligent body construction, and multi-objective optimization, and has significant practical value and promotion potential.

[0084] Applications of the present invention include but are not limited to the following scenarios: Implementing Traffic Guidance and Dynamic Toll Adjustment in Expressway Networks: In addition to the variable speed limits and open hard shoulders involved in traffic flow control, this invention can also be applied to intelligent optimization of traffic guidance and dynamic toll collection schemes in expressway networks. Whether direct traffic guidance or differentiated toll collection, both influence travel demand, thereby impacting the inputs of traffic simulation agents and changing their estimates of the traffic system state. This mechanism, combined with optimization algorithms, can optimize traffic guidance and dynamic toll collection schemes in expressway networks, improving the network's traffic operation.

[0085] Optimization of urban road traffic flow control schemes: In addition to the highway field, the present invention can also be applied to urban road traffic flow control. As long as factors such as signal control are introduced into the sample characteristics, the prediction of urban road traffic operation status can be achieved.

[0086] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0087] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0088] An embodiment of the present invention further provides a processor, which executes a computer program and at least performs the method described above.

[0089] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disc or a read-only optical disc (CD-ROM); the magnetic surface memory can be a magnetic disk memory or a magnetic tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0090] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0091] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0092] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0093] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0094] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0095] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0096] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0097] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0098] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. Those skilled in the art will recognize that, without departing from the scope of the present invention, several equivalent substitutions or obvious variations can be made, and the performance or use of the same should be considered to fall within the scope of protection of the present invention.

Claims

1. A mainline traffic flow control optimization method based on a high-performance traffic simulation agent, characterized in that: The following steps are involved: S1. Select the target study area and determine the regional road network consisting of main line sections and interchange sections; S2. Multi-scenario sample generation based on microscopic traffic simulation: For the target study area, multiple traffic scenarios with different traffic demand levels, road speed limits, and hard shoulder availability are set. Traffic operation status indicators for road sections under each scenario are generated through batch simulation to form a traffic simulation sample set; S3. Build a high-performance traffic simulation agent: Based on the traffic simulation sample set, a deep learning method is used to train a neural network model. The model uses the upstream and downstream topological information and its own attributes of the target road section as input features, and the traffic operation status indicators of the road section as labels to achieve efficient evaluation of the road network traffic status; S4. Optimize the control plan by combining a multi-objective optimization algorithm: take the total travel time and total carbon emissions of the main line as optimization objectives, and the speed limit and hard shoulder opening status of each road section as decision variables, use the traffic simulation agent to evaluate candidate plans, and iterate the optimization algorithm to generate the optimal control plan.

2. The main line traffic flow control optimization method according to claim 1 is characterized in that: In step S2, the generation of multi-scene samples includes: Based on three indicators: route travel volume, road speed limit and hard shoulder open status, different traffic scenarios are randomly set within a preset range. Each scenario is divided into two stages, and the indicators of the two stages are independently and randomly adjusted to simulate the impact of dynamic changes in traffic demand and control measures.

3. The main line traffic flow control optimization method according to claim 1 is characterized in that: In step S3, the method for extracting upstream and downstream topology information includes: Perform multi-layer recursive traversal on the target road section interval, obtain the attributes of its upstream and downstream associated road sections layer by layer, aggregate the features of the associated road sections at each layer, and form a feature vector that represents the spatial dependency relationship.

4. The main line traffic flow control optimization method according to claim 3 is characterized in that: The specific process of the multi-layer recursive traversal includes: a) First-level search: For the target road section, obtain the set of sections of its directly adjacent upstream entry, upstream exit, downstream entry, and downstream exit sections; b) Second-level search: For each road segment interval in the first-level set, recursively search its directly adjacent upstream and downstream road segment intervals to generate a secondary road segment interval set; c) Third-level search: For each road segment interval in the second-level set, continue to recursively search its directly adjacent upstream and downstream road segment intervals to generate a third-level road segment interval set; d) Feature aggregation: The attribute features of the road sections in each level set are aggregated by level to form a multidimensional feature vector that represents the spatial dependency relationship.

5. The main line traffic flow control optimization method according to claim 3 is characterized in that: Aggregating the features of the associated road sections at each level specifically includes: calculating the feature mean of the traffic demand, inherent attributes of the road sections, and control parameters of the road sections in each level set generated by recursive traversal; Among them, the inherent attributes of the road segment include the number of lanes, length and historical traffic volume; Control parameters include speed limit and hard shoulder open status; The mean results of the sets of each level are flattened and concatenated in hierarchical order to form a multi-dimensional feature vector as the input feature of the neural network model.

6. The main line traffic flow control optimization method according to claim 3 is characterized in that: The neural network model is a multi-layer perceptron (MLP), which generates a traffic simulation intelligent agent by learning the mapping relationship between the feature vector and the traffic operation status indicator through training.

7. The main line traffic flow control optimization method according to any one of claims 1 to 6, characterized in that: In step S4, the multi-objective optimization algorithm adopts the genetic algorithm of the NSGA-II framework, iteratively optimizes the population through non-dominated sorting, selection, crossover and mutation operations, and quickly evaluates the fitness of individuals using the traffic simulation agent.

8. The main line traffic flow control optimization method according to any one of claims 1 to 6, characterized in that: In step S4, the total main line travel time is calculated by accumulating the product of the traffic demand and the travel time of each road section; the total carbon emissions are calculated by accumulating the product of the traffic demand, carbon emission factor and road section length of each road section.

9. The main line traffic flow control optimization method according to claim 8, characterized in that: In the calculation of the total travel time, the traffic demand of the road section is obtained by weighted summation of the path travel volume and the path-road section mapping relationship; The calculation of the total carbon emissions specifically includes: calculating the carbon emissions of each vehicle type based on the traffic demand, basic energy consumption, speed correction coefficient, section length and unit energy consumption emission factor of different vehicle types within the road section, and accumulating the results of all vehicle types.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the main line traffic flow control optimization method according to any one of claims 1 to 9 is implemented.

Images