A simulation optimization method and optimization system for active highway control strategy based on Bayesian theory

Abstract

The present invention discloses a simulation optimization method and optimization system for active highway control strategy based on Bayesian theory, which specifically includes building a SUMO simulation environment simulating real traffic flow to dynamically simulate the active highway control strategy; establishing evaluation indicators for traffic network performance; constructing a two-layer programming model, wherein the upper model uses the evaluation index as the optimization target, the lower model uses the balanced traffic flow simulated by SUMO as the decision variable, and simulates the real traffic flow according to the optimization target of the upper model; and iterates and optimizes the decision variables in the two-layer programming model using the Bayesian optimization method to find the best decision variable combination, and the best optimization target can be obtained based on the best decision variable of the combination. The method described in the present invention is applicable to complex highway traffic scenes, and can ensure that the proposed active control strategy can improve the traffic efficiency of the road while ensuring traffic safety, and has low computing requirements.

Description

Technical Field

[0001] The present invention relates to highway control technology, and in particular to a simulation optimization method and optimization system for a highway active control strategy based on Bayesian theory. Background Art

[0002] In recent years, the traffic volume on highways has been increasing, which has brought great challenges to traffic carrying capacity and road management. Therefore, it is of great practical significance to use active and refined traffic control technology to improve the traffic capacity and safety level of key road network sections, realize normal and orderly traffic on highways under normal high-volume situations, and evacuate and rescue in emergency accident scenes.

[0003] In order to save testing time and cost, avoid the dangers of road testing, and improve the efficiency of testing and optimization, digital twins have been widely used in the field of intelligent transportation, and traffic simulation with digital twins as the core has also developed rapidly. Based on traffic simulation technology, road network construction and traffic flow reproduction can be realized, thereby simulating the impact of different active control strategies on traffic flow, providing a scientific quantitative basis for active control strategies.

[0004] In order to design a comprehensive active control collaborative optimization strategy, it is necessary to perform optimization analysis on various strategies. At present, the optimization solution algorithm can be roughly divided into gradient-based algorithm and traditional non-gradient algorithm. The gradient algorithm is limited by the huge amount of computation caused by the derivation in the traffic problem, and the traditional non-gradient algorithm relies on a large-scale search in the whole domain, and its efficiency and accuracy are relatively limited. Therefore, in order to more effectively solve the actual traffic control problem, there is an urgent need for an optimization method with low computational complexity, high solution efficiency and accuracy. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a simulation optimization method and optimization system for active highway control strategies based on Bayesian theory. The method and system are suitable for complex highway traffic scenarios, and can ensure that the proposed active control strategy can improve road traffic efficiency while ensuring traffic safety, and has lower computing requirements.

[0006] Technical solution: To achieve the above purpose, the present invention provides a method for simulating and optimizing the active control strategy of a highway based on Bayesian theory, comprising the following steps:

[0007] Step 1: Build a SUMO simulation environment that can simulate real traffic flow to dynamically simulate the active traffic control strategy of highways;

[0008] Step 2: Establish evaluation indicators on transportation network performance, including operational safety and efficiency;

[0009] Step 3: Construct a two-level planning model, in which the upper model takes the evaluation index as the optimization target, and the lower model is based on the balanced traffic flow simulated by SUMO, with the active traffic control strategy as the decision variable, and simulates the real traffic flow according to the optimization target of the upper model;

[0010] Step 4: Use the Bayesian optimization method to iterate and optimize the decision variables in the bi-level programming model to find the best combination of decision variables, based on which the optimal optimization target can be obtained.

[0011] The method for building a SUMO simulation environment simulating real traffic flow in step 1 is:

[0012] Step 101: importing an open source map database containing actual road conditions into the SUMO simulation environment, and using the netedit plug-in of SUMO to perform localized adjustments on the highway section model in the SUMO simulation environment according to the actual road conditions, so as to ensure that the model is consistent with the actual road conditions;

[0013] Step 102: pre-process the data containing vehicle information in the ETC and gantry systems and import them into the SUMO simulation environment of step 101;

[0014] Step 103: In the SUMO simulation environment, the road attributes of the highway section model are dynamically controlled using the TraCI interface of SUMO to simulate the active highway control strategy.

[0015] Among them, the preprocessing method described in step 102 is: to match the license plates of the data in the ETC and gantry systems to restore the vehicle trajectory, and to clean, fill, and aggregate the data with missing, redundant, and disordered information and convert them into traffic information that can be recognized by the SUMO simulation environment.

[0016] The active traffic control strategy described in step 1 includes:

[0017] Variable speed limit: Dynamically adjust speed limit signs based on traffic flow and road conditions to optimize vehicle operating speed;

[0018] Ramp control: Regulate the number or type of vehicles entering the highway by controlling the entrance and exit of the ramp;

[0019] Emergency Lane Control: Controls the opening of emergency lanes to vehicles to ease traffic congestion during peak hours.

[0020] Among them, the operational safety described in step 2 refers to using the instantaneous speed standard deviation as an indicator to measure the stability and safety of the road network. The smaller the standard deviation value, the fewer acceleration and deceleration of vehicles on the highway, that is, the better the road network stability and safety; operational efficiency refers to using the average travel time as an indicator to measure the overall operational efficiency of the road network. The smaller the average travel time value, the smoother the vehicle travel on the highway.

[0021] Among them, the method for establishing evaluation indicators for traffic network performance described in step 2 is: when evaluating traffic network performance, first standardize indicators of different magnitudes including instantaneous speed standard deviation and average travel time, and the standardized indicator values ​​are within the set numerical range; then, weighted summation of the standardized indicator values ​​is performed according to the set weights to obtain a comprehensive evaluation value to measure the importance of each indicator.

[0022] The balanced traffic flow in step 3 means that after setting the starting and ending points for each vehicle in the SUMO simulation environment, the vehicles are evenly distributed and placed in the road network;

[0023] The decision variables include continuous variables and discrete variables; the continuous variables include the set value of the road speed limit to adjust the vehicle running speed, and the discrete variables include the lane restriction regulations, that is, the number of vehicles or the type of vehicles entering the highway are adjusted through ramp and emergency lane control.

[0024] The method of iterating and optimizing the decision variables in the bi-level programming model using the Bayesian optimization method in step 4 to find the best combination of decision variables includes:

[0025] Step 401: constructing a proxy model of an objective function and constraint conditions based on a Gaussian process, wherein the objective function refers to the optimization target of the upper model in the two-level programming model, and the constraint conditions refer to the decision variables of the lower model in the two-level programming model;

[0026] Step 402: Iterate the proxy model through the Bayesian optimization method to optimize the decision variables of the lower model, so as to find the optimal value of one or more decision variables to minimize the objective function.

[0027] The Gaussian kernel function formula of the proxy model in step 401 is as follows:

[0028]

[0029] Among them, σ and l are hyperparameters, i and j represent different sampling rounds, and x i 、x j is the sampling point, which refers to the point of the decision variable.

[0030] The iterative optimization process in step 402 is specifically as follows:

[0031] Step 4021: Before iteration, a number of initial sampling points are randomly selected using the Latin hypercube sampling method to perform a multi-starting point search to ensure that the sampling is approximately random in the multivariate parameter distribution and avoid falling into a local minimum;

[0032] Step 4022: Use the log-marginal likelihood method of the maximum Gaussian process to estimate the optimal hyperparameters for each iteration to improve the prediction accuracy of the proxy model. The specific process is as follows:

[0033] y|σ,l~N(0,K yy (σ,l));

[0034] Among them, σ and l are Gaussian model hyperparameters; y is the function estimate, that is, the comprehensive evaluation value at this sampling point, which has a mean of 0 and a variance of K. yy (σ,l) Gaussian distribution;

[0035]

[0036] is the covariance matrix, where n represents the sampling round, x n represents the nth sampling point;

[0037]

[0038] Among them, L is the maximum likelihood estimate, K yy (σ,l) -1 is the inverse matrix of the covariance matrix;

[0039] Step 4023: Use the acquisition function to select a new sampling point. The acquisition function selects a new sampling point x based on the confidence interval upper bound algorithm. n+1 , the acquisition function is based on the estimated value and estimation error of the objective function, and uses the weight hyperparameter β n A balance is made between the estimated value and the estimation error to minimize the acquisition function H(x) to find new sampling points that can improve the model accuracy and reduce the optimization error:

[0040]

[0041] Step 4024: Set the new sampling point x n+1 Set up the simulation in the SUMO simulation environment and obtain the function estimation value at the sampling point based on the evaluation index, that is, the comprehensive evaluation value y n+1 , based on (x n+1 ,y n+1 )Update the proxy model;

[0042] Step 4025: Repeat steps 4023 and 4024 until the iteration reaches the termination condition and stops. At this time, the optimal sampling points are obtained to form the best decision variable combination.

[0043] The present invention provides a highway active control strategy simulation optimization system based on Bayesian theory, comprising the following modules:

[0044] Simulation environment building module: used to build a SUMO simulation environment that simulates real traffic flow to dynamically simulate the active control strategy of highway traffic;

[0045] Evaluation index establishment module: used to establish evaluation indicators on transportation network performance, including operational safety and operational efficiency;

[0046] Two-layer planning model construction module: used to construct a two-layer planning model, in which the upper-layer model uses the evaluation index as the optimization target, and the lower-layer model is based on the balanced traffic flow simulated by SUMO, with the active traffic control strategy as the decision variable, and simulates the real traffic flow according to the optimization target of the upper-layer model;

[0047] Optimal decision variable screening module: The Bayesian optimization method is used to iterate and optimize the decision variables in the bi-level programming model to find the best combination of decision variables, based on which the optimal optimization target can be obtained.

[0048] Beneficial effects: The present invention has the following advantages: 1. The present invention constructs a Gaussian process proxy model to approximate the objective function in a complex optimization problem, thereby avoiding the huge amount of calculation caused by direct derivation of the existing gradient algorithm, thereby effectively processing complex problems and reducing the amount of calculation;

[0049] 2. The present invention uses the Bayesian optimization method to search for the optimal active control strategy. In the search process, the sampling function is combined with the previous sampling results and the prior knowledge of the problem to guide the selection of new sampling points, so as to achieve a good balance between exploration and utilization, which can avoid wasting resources in the explored areas and ensure the discovery of better solutions, thereby improving the efficiency of the optimization process and the accuracy of the results. The method and system of the present invention are suitable for solving problems that are difficult to model, complex in changes, or have many unknown factors;

[0050] 3. Based on the SUMO simulation environment, the present invention constructs a standardized framework system for highway scenarios and active control strategies. The system takes safety and efficiency as dual optimization goals, and finds the optimal solution through the Bayesian optimization algorithm, thereby providing a basis for the qualitative combination and quantitative analysis of active control strategies. This method ensures that the proposed active control strategy can not only ensure the safety of traffic flow, but also improve the traffic efficiency of highways. It has the advantages of strong practicality and high compatibility, and is suitable for controlling complex highway traffic scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the simulation road network framework of the present invention;

[0052] Figure 2 Schematic diagram of the process of this method. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described in detail below in conjunction with the embodiments and drawings.

[0054] like Figure 1 As shown, the present invention focuses on the highway interchange scenario and selects an interchange connecting three directions for simulation. The highway interchange realizes the interconnection between highways and is also a key node connecting the entrances and exits of ordinary roads. In the highway interchange interweaving area, the traffic conflict between the main road and the ramp is an important cause of traffic congestion and accidents. The quantitative study of ramp control strategy has important research significance for the optimization of highway traffic organization. The present invention proposes a simulation optimization method for active highway control strategy based on Bayesian theory, which can provide scientific quantitative methods and systems for common highway control strategies such as ramp control, emergency lane control, and variable speed limit, as follows.

[0055] like Figure 2 As shown, the present invention discloses a method for simulating and optimizing active highway control strategies based on Bayesian theory, comprising the following steps:

[0056] Step 1: Build a SUMO simulation environment and restore the real traffic flow on the highway in the SUMO simulation environment based on ETC and gantry system data, including:

[0057] Step 1.1: SUMO simulation environment construction and modification

[0058] a) Import highway sections through the open source map database OpenStreetMap and convert them into SUMO simulated road network files (*.net.xml);

[0059] b) Refer to map platforms such as Amap and Baidu, and use SUMO's netedit plug-in to make localized fine modifications based on actual road conditions.

[0060] Step 1.2: Preprocessing and importing traffic data

[0061] a) Perform license plate matching on ETC and gantry system data to restore vehicle trajectory;

[0062] b) To address the possible problems of missing, redundant, disordered data, etc., pre-processing operations such as data cleaning, filling and aggregation are performed according to the rationality of vehicle travel time and space;

[0063] c) Convert the preprocessed traffic flow information into a traffic flow format that can be recognized in SUMO simulation.

[0064] Step 1.3: Simulate and reproduce common active control strategies on highways

[0065] a) Select active highway control strategies suitable for testing in simulation scenarios, such as variable speed limits, ramp and emergency lane control, etc.;

[0066] b) Use the TraCI interface provided by SUMO to dynamically control road attributes to simulate active control strategies, such as modifying road speed limits and the types of vehicles allowed to pass.

[0067] Step 2: Establish an evaluation system that takes into account both safety and efficiency, including:

[0068] a) The average travel time is used as an indicator to measure the overall operating efficiency of the road network. After the simulation is completed, SUMO simulation will output the average travel time of all vehicles from the starting point to the end point during the entire simulation period. The smaller the value, the smoother the vehicle travels on the road;

[0069] b) The instantaneous speed standard deviation is used as an indicator to measure the stability and safety of the road network. The larger the speed standard deviation, the more frequent the acceleration and deceleration of vehicles, the worse the stability and the less safe it is;

[0070] c) The speed information of all vehicles on a specific road section can be obtained through the TraCI interface. Since the average speed is difficult to describe the discrete degree of speed, the instantaneous speed standard deviation of all sections is calculated every 5 minutes and then averaged over the entire period to represent safety;

[0071] d) Since the evaluation indicators can be obtained accurately at the road section level, vehicle congestion and safety hazards can also be located at the road section or even lane level;

[0072] e) Since there is a difference in magnitude between the average driving time and the standard deviation of the instantaneous speed, they need to be standardized and then weighted summed. The specific weight depends on the tendency and balance between safety and efficiency.

[0073] Step 3: Construct a two-level planning model based on traffic simulation

[0074] a) The upper model takes the evaluation index that takes into account both road network safety and efficiency in step 2 as the optimization target;

[0075] b) The lower model is a balanced traffic flow based on SUMO simulation. After setting the starting and ending points for the vehicles, the dynamic traffic allocation function built into SUMO is used to evenly distribute the vehicles in the road network;

[0076] c) Taking specific active traffic control measures as decision variables, including continuous variables (such as the set value of road speed limit) and discrete variables (such as lane restrictions);

[0077] d) In terms of constraints, the values ​​of decision variables must obey objective laws (such as the speed limit cannot be negative and the lane opening and closing is a 0-1 variable), and must also refer to reality and regulations and be based on prior knowledge to avoid meaningless calculations (such as the speed limit should not be too high or too low).

[0078] Step 4: Establish a solution method based on Bayesian optimization to iterate and optimize the traffic active control strategy, including:

[0079] Step 4.1: Build a surrogate model based on Gaussian process

[0080] a) For simulation models that cannot be differentiated, a proxy model is constructed based on Gaussian process to approximate the objective function;

[0081] b) The model uses a Gaussian kernel function whose hyperparameters can be dynamically adjusted, where σ and l are hyperparameters, i and j represent different sampling rounds, and x i 、x j is the specific sampling point, i.e., the point that refers to the decision variable;

[0082]

[0083] c) Use the log-marginal likelihood method of the maximum Gaussian process to estimate the optimal hyperparameters for each iteration, which are used to estimate the hyperparameters in the proxy model. These hyperparameters determine the model's predictive power and uncertainty estimates. By optimizing the hyperparameters, the proxy model's approximation of the true target function can be improved, thereby making more accurate predictions in subsequent sampling and optimization. ;

[0084] y|σ,l~N(0,K yy (σ,l))

[0085] Among them, σ and l are Gaussian model hyperparameters, y is the function value, which has a mean of 0 and a variance of K yy (σ,l) Gaussian distribution;

[0086]

[0087] is the covariance matrix, where n represents the nth sampling point, and the kernel function calculation method is as described above;

[0088]

[0089] Among them, L is the maximum likelihood estimate, K yy (σ,l) -1 is the inverse of the covariance matrix.

[0090] Step 4.2: Multi-start random search

[0091] a) In order to avoid falling into the local minimum, before iteration, it is necessary to randomly select several initial points for multi-starting point search, and then update these sampling points through an iterative process. Each iteration involves using the Gaussian process as a proxy model to predict the possible objective function value of the new sampling point;

[0092] b) Latin hypercube sampling is used to ensure that the sampling is approximately random in the multivariate parameter distribution.

[0093] Step 4.3: The acquisition function selects new sampling points

[0094] a) The purpose of the acquisition function is to determine which point should be selected for the next sampling in order to most hopefully improve the final optimization result. This function selects the confidence interval upper bound algorithm and comprehensively considers the estimated value μ n (x) (i.e., the posterior mean) and the estimated error σ n (x) (i.e., the posterior variance) is used as the basis for selecting the next sampling point, using the weight hyperparameter β n A balance is struck between the estimated value and the estimation error to minimize the acquisition function H(x):

[0095]

[0096] b) As the number of iterations and sampling points increases, the importance of existing sampling points also increases, and the estimated variance σ used for exploration n The importance of (x) also gradually decreases, that is, the weight parameter β is set n , where n represents the iteration round, s represents the initial multi-starting point sampling number, and α is the exponential function hyperparameter;

[0097]

[0098] Step 4.4: Iterative update of proxy model

[0099] a) Set the new sampling point x of the active control strategy n+1 Set up in simulation;

[0100] b) After running the simulation, the comprehensive evaluation value y is obtained based on the evaluation system in step 2.1 n+1 ;

[0101] c)(x n+1 ,y n+1 ) is added to the historical sampling points and the Gaussian process-based proxy model is updated.

[0102] Step 4.5: Repeat steps 4.3 and 4.4 until the iteration reaches the termination condition:

[0103] a) The number of iterations reaches 30;

[0104] b) The condition number of the Gaussian kernel function variance matrix exceeds 10 16 , which means that the sampling points are highly clustered and the sampling is sufficient; finally, the optimal decision variable combination x is obtained opt , as a quantitative basis for active management and control strategies.

[0105] The present invention provides a highway active control strategy simulation optimization system based on Bayesian theory, comprising the following modules:

[0106] Simulation environment building module: used to build a SUMO simulation environment that simulates real traffic flow to dynamically simulate the active control strategy of highway traffic;

[0107] Evaluation index establishment module: used to establish evaluation indicators on transportation network performance, including operational safety and operational efficiency;

[0108] Two-layer planning model construction module: used to construct a two-layer planning model, in which the upper-layer model uses the evaluation index as the optimization target, and the lower-layer model is based on the balanced traffic flow simulated by SUMO, with the active traffic control strategy as the decision variable, and simulates the real traffic flow according to the optimization target of the upper-layer model;

[0109] Optimal decision variable screening module: The Bayesian optimization method is used to iterate and optimize the decision variables in the bi-level programming model to find the best combination of decision variables, based on which the optimal optimization target can be obtained.

[0110] The method and system described in the present invention focus on the active control strategy of highways, simulate the impact of active control strategies on traffic flow based on SUMO simulation, build an evaluation system that takes into account both safety and efficiency, and use a solution system based on Bayesian optimization to iterate and optimize the active traffic control strategy. The present invention focuses on building a standardized system framework with high compatibility, and can couple multiple active control strategies for collaborative optimization, providing a basis for qualitative combination and quantitative analysis of strategies.

Claims

1. A simulation optimization method for active highway control strategy based on Bayesian theory, characterized in that: The following steps are involved: Step 1: Build a SUMO simulation environment that simulates real traffic flow to dynamically simulate the active traffic control strategy of highways; Step 2: Establish evaluation indicators on transportation network performance, including operational safety and efficiency; Step 3: Construct a two-level planning model, in which the upper model takes the evaluation index as the optimization target, and the lower model is based on the balanced traffic flow simulated by SUMO, with the active traffic control strategy as the decision variable, and simulates the real traffic flow according to the optimization target of the upper model; Step 4: Use the Bayesian optimization method to iterate and optimize the decision variables in the bi-level programming model to find the best combination of decision variables, based on which the optimal optimization target can be obtained.

2. The Bayesian-based highway active control strategy simulation optimization method according to claim 1 is characterized in that: The method for building a SUMO simulation environment simulating real traffic flow in step 1 is: Step 101: importing an open source map database containing actual road conditions into the SUMO simulation environment, and using the netedit plug-in of SUMO to perform localized adjustments on the highway section model in the SUMO simulation environment according to the actual road conditions, so as to ensure that the model is consistent with the actual road conditions; Step 102: pre-processing the data containing vehicle information in the ETC and gantry systems and importing them into the SUMO simulation environment of step 101; the pre-processing method is: performing license plate matching on the data in the ETC and gantry systems to restore the vehicle trajectory, and cleaning, filling, and aggregating the data with missing, redundant, and disordered information and converting them into traffic information that can be recognized by the SUMO simulation environment; Step 103: In the SUMO simulation environment, the road attributes of the highway section model are dynamically controlled using the TraCI interface of SUMO to simulate the active highway control strategy.

3. The method for simulating and optimizing the active highway control strategy based on Bayesian theory according to claim 1 is characterized in that: The active traffic control strategy described in step 1 includes: Variable speed limit: Dynamically adjust speed limit signs based on traffic flow and road conditions to optimize vehicle operating speed; Ramp control: Regulate the number or type of vehicles entering the highway by controlling the entrance and exit of the ramp; Emergency Lane Control: Controls the opening of emergency lanes to vehicles to ease traffic congestion during peak hours.

4. The method for simulating and optimizing the active highway control strategy based on Bayesian theory according to claim 1 is characterized in that: The operational safety described in step 2 refers to using the instantaneous speed standard deviation as an indicator to measure the stability and safety of the road network. The smaller the standard deviation value, the less acceleration and deceleration of vehicles on the highway, that is, the better the road network stability and safety; the operational efficiency refers to using the average travel time as an indicator to measure the overall operational efficiency of the road network. The smaller the average travel time value, the smoother the vehicle travel on the highway.

5. The method for simulating and optimizing the active highway control strategy based on Bayesian theory according to claim 4 is characterized in that: The method for establishing evaluation indicators for traffic network performance described in step 2 is: when evaluating traffic network performance, first standardize indicators of different magnitudes including instantaneous speed standard deviation and average travel time, and the standardized indicator values ​​are within the set numerical range; then, weighted summation of the standardized indicator values ​​is performed according to the set weights to obtain a comprehensive evaluation value to measure the importance of each indicator.

6. The Bayesian-based highway active control strategy simulation optimization method according to claim 1 is characterized in that: The balanced traffic flow in step 3 means setting the starting and ending points for each vehicle in the SUMO simulation environment and evenly distributing the vehicles in the road network; The decision variables include continuous variables and discrete variables; The continuous variables include the set value of the road speed limit to adjust the vehicle running speed, and the discrete variables include the lane restrictions, that is, the number or type of vehicles entering the highway is adjusted through ramp and emergency lane control.

7. The method for simulating and optimizing active highway control strategies based on Bayesian theory according to claim 1 is characterized in that: The method described in step 4 of iterating and optimizing the decision variables in the bi-level programming model using the Bayesian optimization method to find the best combination of decision variables includes: Step 401: constructing a proxy model of an objective function and constraint conditions based on a Gaussian process, wherein the objective function refers to the optimization target of the upper model in the two-level programming model, and the constraint conditions refer to the decision variables of the lower model in the two-level programming model; Step 402: Iterate the proxy model through the Bayesian optimization method to optimize the decision variables of the lower model, so as to find the optimal value of one or more decision variables to minimize the objective function.

8. The method for simulating and optimizing the active highway control strategy based on Bayesian theory according to claim 7 is characterized in that: The Gaussian kernel function formula of the proxy model in step 401 is as follows: ； in, and is a hyperparameter, and Represents different sampling rounds, , is the sampling point, which refers to the point of the decision variable.

9. The method for simulating and optimizing active highway control strategies based on Bayesian theory according to claim 8 is characterized in that: The iterative optimization process described in step 402 is specifically as follows: Step 4021: Before iteration, a number of initial sampling points are randomly selected using the Latin hypercube sampling method to perform a multi-starting point search to ensure that the sampling is approximately random in the multivariate parameter distribution and avoid falling into a local minimum; Step 4022: Use the log-marginal likelihood method of the maximum Gaussian process to estimate the optimal hyperparameters for each iteration to improve the prediction accuracy of the proxy model. The specific process is as follows: ； in, and is the Gaussian model hyperparameter; is the function estimate, that is, the comprehensive evaluation value at the sampling point, which has a mean of 0 and a variance of Gaussian distribution; ； is the covariance matrix, where represents the sampling round, Representative sampling points; ； in, is the maximum likelihood estimate, is the inverse matrix of the covariance matrix; Step 4023: Select new sampling points using the acquisition function, which selects new sampling points based on the confidence interval upper bound algorithm. , the acquisition function is based on the estimated value and estimated error of the objective function, and uses the weight hyperparameter A balance is made between the estimated value and the estimation error to minimize the acquisition function , thereby finding new sampling points that can improve model accuracy and reduce optimization errors: ； in, represents the estimated value, that is, the posterior mean, represents the estimation error, i.e., the posterior variance; Step 4024: Set the new sampling point Set up the simulation in the SUMO simulation environment and obtain the function estimation value at the sampling point based on the evaluation index, that is, the comprehensive evaluation value ,based on Update proxy model; Step 4025: Repeat steps 4023 and 4024 until the iteration reaches the termination condition and stops. At this time, the optimal sampling points are obtained to form the best decision variable combination.

10. A simulation optimization system for active highway control strategy based on Bayesian theory, characterized in that: Includes the following modules: Simulation environment building module: used to build a SUMO simulation environment that simulates real traffic flow to dynamically simulate the active control strategy of highway traffic; Evaluation index establishment module: used to establish evaluation indicators on transportation network performance, including operational safety and operational efficiency; Two-layer planning model construction module: used to construct a two-layer planning model, in which the upper-layer model uses the evaluation index as the optimization target, and the lower-layer model is based on the balanced traffic flow simulated by SUMO, with the active traffic control strategy as the decision variable, and simulates the real traffic flow according to the optimization target of the upper-layer model; Optimal decision variable screening module: The Bayesian optimization method is used to iterate and optimize the decision variables in the bi-level programming model to find the best combination of decision variables, based on which the optimal optimization target can be obtained.

Images