Traffic signal timing method and system based on agent-assisted hierarchical optimization

Abstract

The application provides a traffic signal timing method and system based on agent-assisted hierarchical optimization, and belongs to the technical field of urban traffic signal management. Through the construction of a hierarchical joint optimization framework, the synergistic effect of global exploration and local development is realized, and the bottleneck that the traditional method is easy to fall into local optimum or slow convergence in the later optimization stage is broken. In the global stage, the efficiency of the RBF model and the improved search strategy can quickly locate the promising area in the wide solution space; in the local stage, the uncertainty quantification ability of the GP model and the strong local optimization ability of DE are used to carry out deep mining in the identified high-quality area. The two complement each other, and the traffic signal timing scheme with higher quality (the average delay time is significantly reduced) and faster convergence speed can be obtained.

Description

Technical Field

[0001] This invention relates to the field of urban traffic signal management technology, specifically to a traffic signal timing method and system based on agent-assisted hierarchical optimization. Background Technology

[0002] Urban road intersections are the core and challenging nodes in urban traffic management and control. Traffic signal control, as the core technical means of intersection traffic management, can effectively guide traffic flow at intersections and alleviate road network congestion by dynamically adjusting the timing of vehicle passage in each approach lane. It is a key technical support for improving urban traffic efficiency.

[0003] From a control logic perspective, current mainstream traffic signal control technologies mainly include three types: timing control, inductive control, and adaptive control. To overcome the limitations of traditional TSC technologies, various metaheuristic algorithms, with their advantages of no gradient dependency and strong global search capabilities, have become the mainstream research direction in the field of traffic signal optimization and are currently the closest existing implementation system to the technical solution of this invention. These algorithms can break free from the constraints of traditional analytical models and efficiently solve for the optimal signal timing parameters in complex traffic scenarios, providing a new technical path for multi-intersection, networked TSC optimization.

[0004] Among existing similar technical solutions, Particle Swarm Optimization (PSO) is one of the most widely used metaheuristic algorithms. Besides PSO, classic metaheuristic algorithms such as Genetic Algorithm (GA) and Differential Evolutionary Algorithm (DE) are also widely applied in TSC optimization scenarios, forming a variety of mature existing approximate technical solutions. In the application of Genetic Algorithms, existing research improves timing performance through optimization algorithm iteration mechanisms. Kong Xianming et al. improved the traditional GA by adding elite retention and population elimination mechanisms. Simulation verification shows that this improved scheme can effectively optimize intersection timing strategies and improve the overall traffic capacity of intersections. In the application of Differential Evolutionary Algorithms, existing research has constructed intersection signal optimization models based on the DE algorithm. By comparing with two traditional schemes—fixed timing control and vehicle-sensing dynamic control—it has been confirmed that the DE algorithm can effectively reduce average vehicle delay and optimize traffic flow. Meanwhile, various novel swarm intelligence optimization algorithms are also gradually being applied in the TSC field, enriching the existing similar technical system.

[0005] In recent years, the industry has gradually developed a surrogate model-assisted traffic signal optimization framework, which has become a mainstream cutting-edge technology for efficient optimization of TSC in large-scale road networks and an important existing technology system that fits the application scenario of this invention. The core idea of ​​this technology is to use a low-computational-cost surrogate model to replace some of the time-consuming real traffic simulation evaluations. By constructing a high-precision surrogate response surface, it quickly searches for optimal timing candidate solutions, significantly reducing the frequency of simulation calls during iterative optimization, and significantly improving overall computational efficiency while ensuring optimization effects. Based on the construction principle and implementation method of the surrogate model, existing mainstream technical solutions can be divided into three categories: physical surrogate models, regression surrogate models, and analytical surrogate models. The first category is the physical surrogate model optimization method. This type of method relies on traffic physics to construct a simplified mechanism model, replacing the high-precision microscopic simulation model to complete performance evaluation, effectively reducing computational overhead. The second category is the regression surrogate model optimization method. This type of method relies on machine learning regression models to fit the mapping relationship between timing parameters and traffic evaluation indicators, achieving rapid prediction of candidate solution performance, and is suitable for large-scale and expensive optimization scenarios. The third category is the analytical surrogate model optimization method. This type of method filters out inferior candidate solutions through data analysis, feature selection, and model prediction mechanisms, and only performs real simulation evaluation on high-quality solutions, thus accurately reducing the amount of invalid computation.

[0006] However, traditional signal timing optimization methods struggle to achieve global optima. While modern intelligent optimization algorithms possess powerful search capabilities, they require performance evaluation of numerous candidate signal timing schemes when solving traffic signal control optimization problems. This evaluation process typically relies on high-fidelity microscopic traffic simulations, which are time-consuming and computationally expensive, resulting in an excessively long overall optimization time, making them unsuitable for practical engineering applications. Secondly, while some existing surrogate-assisted optimization methods reduce the number of simulations by constructing approximate models, most employ a single global surrogate model. This static strategy cannot adapt to the dynamic changes in population distribution and optimization objectives during evolution, easily leading to accumulated model prediction biases, misleading the search direction, and causing the algorithm to prematurely converge to suboptimal solutions—essentially falling into "local optima." Furthermore, it exhibits poor adaptability to large-scale, high-dimensional road network optimization problems. In summary, the shortcomings of existing technologies are as follows: when dealing with the optimization problem of traffic signal control in large-scale urban road networks, the single agent model strategy is difficult to adapt to the dynamically changing search environment, and the optimization efficiency and accuracy are limited; the algorithm is prone to getting trapped in local optima, and there is an imbalance between global search and local development capabilities; it relies on a large number of expensive simulation evaluations, resulting in high computational costs and making it difficult to obtain high-quality globally optimal signal timing schemes within a limited computational budget. Summary of the Invention

[0007] The purpose of this invention is to provide a traffic signal timing method and system based on agent-assisted hierarchical optimization, which significantly reduces the number of expensive simulation evaluations while ensuring that the optimization algorithm can efficiently and stably search for the globally optimal traffic signal timing scheme, thereby solving at least one of the technical problems existing in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a traffic signal timing method based on agent-assisted hierarchical optimization, comprising:

[0010] Collect real traffic flow information from the urban road network and obtain phase information from traffic lights;

[0011] The timing parameters of each phase are combined with the traffic light phase information and encoded to form a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm.

[0012] An improved artificial bee colony optimization algorithm is used to perform iterative search using an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes through the filling criterion for simulation verification, guiding the search to converge toward the optimal region.

[0013] The optimized timing scheme is input into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, the system simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. The real traffic system is decomposed into two categories: inherent characteristics and dynamic characteristics, completing a digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. Both types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to a high-fidelity simulation system.

[0014] The road network evaluation index output by the SUMO micro-simulation platform is used as real performance feedback to update the RBF proxy model and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

[0015] As a further limitation of the first aspect of the present invention, the artificial bee colony algorithm assisted by the RBF surrogate model includes: firstly, generating N candidate timing schemes through Latin hypercube sampling, each candidate timing scheme obtaining its precise fitness value through evaluation, these evaluated schemes and their prediction function values ​​constituting an initial database DB; the number of evaluations NFEs is initialized to N, N optimal samples are selected from the DB to train the RBF surrogate model, and the improved ABC algorithm is used to search using this model, sequentially executing the improved hired bee stage, determining whether to enter the improved observation bee stage based on probability p, determining whether to trigger the re-initialization of scout bees based on trial constraints, and locating the optimal individual through the filling criterion for real simulation evaluation to update the DB; then entering the local search stage, selecting N optimal samples from the DB to train the GP surrogate model, establishing a local optimal region, using the DE / current-to-best / 1 operator for local search, and then locating the optimal individual through the LCB filling criterion for real evaluation and updating the DB; finally, determining whether the maximum number of function evaluations MaxFEs has been reached, if not, returning to the global search stage to continue iteration, otherwise terminating the algorithm and outputting the final result.

[0016] As a further limitation of the first aspect of this invention, the search strategy of the ABC algorithm is improved in a layered manner. This improved search strategy is applied to the hired bee stage, relying on a novel update mechanism that introduces random variables to enhance the global exploration capability of the population, efficiently uncovering untraversed high-quality regions in the solution space, and providing a high-quality initial population for subsequent local refined searches. The improved search strategy is as follows: ;in, As the perturbation weight, it is a random number between [-1, 1]; , They are distinct random integers between (0, N); This represents the green light duration for the j-th phase of the i-th test timing scheme; , This represents the green light duration of the j-th phase from two parent timing schemes randomly selected from the population.

[0017] As a further limitation of the first aspect of this invention, a local search strategy is introduced into the ABC algorithm to specifically enhance its local optimization performance, enabling rapid convergence to the precise location of the optimal solution and achieving an effective balance between global wide-area search and local fine-grained optimization; wherein, the local search strategy is as follows: ;in, This represents the j-th dimension component of the current best individual. As the perturbation weight, it is a random number between [-1, 1]; N is the population size, and in signal timing, the population refers to the number of candidate timing schemes; Let represent the j-th dimension component of the k-th individual randomly selected from the population.

[0018] As a further limitation of the first aspect of the present invention, the specific steps of applying the GP model to the local search stage of the algorithm include: First, selecting N optimal timing schemes with the minimum average delay from the database DB, calculating the range of green light duration for each phase in the population, and adaptively determining the lower bound LB and upper bound UB of the local search for each phase with the duration of each phase of the current optimal timing scheme as the center, constructing the optimal local region, and focusing the search space on the neighborhood of high-potential timing schemes; Second, selecting N optimal samples from the DB to train a local GP surrogate model, establishing a mapping from timing schemes to average delay, so that the model can predict the mean delay and uncertainty variance of candidate schemes; Third, generating NP candidate timing schemes as the initial population within the local optimal region through Latin hypercube sampling, and using the trained GP model to quickly predict the delay performance of each scheme, characterizing the fitness distribution of the local solution space; Fourth, for each... The individual is assigned a scaling factor F and a crossover probability CR for differential evolution. Experimental timing schemes are generated using the DE / current-to-best / 1 operator. The mutation direction is guided by the phase duration of the current optimal scheme. The step size is controlled by the scaling factor. After binomial crossover, new candidate schemes are generated, and their fitness is evaluated using the GP model. In the fifth step, the GP predicted delays of the experimental schemes are compared with those of the parent schemes, and the high-quality timing schemes with smaller average delays are retained for the next generation. In the sixth step, the mean predicted delay and uncertainty variance of each candidate scheme are comprehensively considered using the lower confidence bound (LCB) criterion, prioritizing high-potential timing schemes with low predicted delays and high uncertainty. In the seventh step, the optimal potential scheme selected by the LCB criterion is evaluated using SUMO real traffic simulation to obtain accurate average delays. The timing scheme and its actual performance are stored in the database DB, the NFEs count is updated, and this round of local search iteration is completed.

[0019] As a further limitation of the first aspect of the present invention, the DE / current-to-best / 1 mutation operator is selected to complete the local population update. This operator uses the best individual in the current population as the mutation reference vector, and regulates the direction of population iteration mutation through the guiding role of the best individual. It can maintain population diversity, avoid premature convergence, and ensure a stable convergence speed of the algorithm, thereby achieving a dynamic balance between exploration and development capabilities. The specific update mechanism is as follows:

[0020]

[0021]

[0022] in, Represents the best individual in the current population. It is a control factor. , Two different individuals randomly selected from the population, Represents the individuals after the crossover. Represents the parent generation; It is a uniformly distributed random number. Let represent the crossover probability in the j-th dimension.

[0023] In a second aspect, the present invention provides a traffic signal timing system based on agent-assisted hierarchical optimization, comprising:

[0024] The acquisition module is used to acquire real traffic flow information and traffic light phase information of the urban road network;

[0025] The encoding module is used to combine the timing parameters of each phase with the traffic light phase information to encode a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm.

[0026] The iterative search module is used to perform iterative search using an improved artificial bee colony optimization algorithm assisted by an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes for simulation verification through the filling criterion, guiding the search to converge toward the optimal region.

[0027] The simulation module is used to input the optimized timing scheme into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, it simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. Specifically, it decomposes the real traffic system into two categories: inherent characteristics and dynamic characteristics, completing the digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. Both types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to a high-fidelity simulation system.

[0028] The update module is used to take the road network evaluation index output by the SUMO micro-simulation platform as real performance feedback, update the RBF proxy model, and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

[0029] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the traffic signal timing method based on agent-assisted hierarchical optimization as described in the first aspect.

[0030] Fourthly, the present invention provides a computer device including a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions executable by the processor, and the processor calling the program instructions to execute the traffic signal timing method based on agent-assisted hierarchical optimization as described in the first aspect.

[0031] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the traffic signal timing method based on agent-assisted hierarchical optimization as described in the first aspect.

[0032] Terminology Explanation:

[0033] (1) Surrogate model: A low-cost approximation model used to simulate high-cost, precise simulation evaluations. By training on historical data, the performance of new candidate solutions can be quickly predicted. Commonly used models include radial basis functions, Gaussian processes, etc.

[0034] (2) Expensive optimization problems: Problems where the evaluation of the objective function requires huge computational resources or time, such as the need to run a complete traffic simulation to evaluate a signal timing scheme in this invention.

[0035] (3) Radial basis function: a real-valued function with distance as the independent variable. Its linear combination can construct a surrogate model. It has the characteristics of simple structure, high computational efficiency and strong nonlinear fitting ability.

[0036] (4) Gaussian process: a nonparametric probabilistic model based on the Bayesian framework that can not only predict function values ​​but also quantify the uncertainty of predictions, and is used to guide the balance between exploration and development in the search process.

[0037] (5) Filling Criterion: In surrogate-assisted optimization, the strategy used to determine which candidate solution predicted by the surrogate model should be sent to the expensive real function for evaluation.

[0038] (6) Hierarchical optimization framework: The optimization process is decomposed into multiple levels or stages, and each level adopts a different strategy or model (such as global search and local refinement) to achieve an overall framework for collaborative optimization.

[0039] The beneficial effects of this invention are as follows: By constructing a hierarchical joint optimization framework, it achieves synergistic effects between global exploration and local development, overcoming the bottleneck of traditional methods that easily get trapped in local optima or have slow convergence in the later stages of optimization. Specifically, the global stage utilizes the efficiency of the RBF model and the improved search strategy to quickly locate promising regions in a broad solution space; the local stage leverages the uncertainty quantification capability of the GP model and the powerful local optimization capability of the DE model to conduct in-depth mining within the identified high-quality regions. These two aspects complement each other, enabling this invention to obtain higher-quality traffic signal timing schemes (significantly reduced average delay time) while achieving faster convergence speed within a limited and expensive simulation budget. Experiments demonstrate that on a real urban road network dataset, the HSAABC algorithm proposed in this invention reduces the average road network delay time by an average of 9.35% compared to the existing mainstream SAEA algorithm, exhibiting stronger global optimization capability, faster convergence speed, and better robustness.

[0040] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating the urban road network traffic signal optimization method according to an embodiment of the present invention.

[0043] Figure 2 This is a simulation framework diagram of urban road network traffic signals according to an embodiment of the present invention.

[0044] Figure 3 This is a simulation framework diagram for evaluating the performance of urban road network traffic signal optimization according to an embodiment of the present invention.

[0045] Figure 4 This is a flowchart of the hierarchical artificial bee colony algorithm based on agent assistance as described in an embodiment of the present invention. Detailed Implementation

[0046] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0047] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0048] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0049] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0050] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0051] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0052] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0053] The purpose of this invention is to provide a traffic signal timing method based on surrogate-assisted hierarchical optimization. By constructing a hierarchically improved surrogate-assisted artificial bee colony algorithm (HSAABC), a radial basis function (RBF) surrogate model is used for efficient exploration in the global search phase, and a Gaussian process (GP) surrogate model is used to guide the differential evolution (DE) algorithm for depth search in the local refinement phase. This method aims to achieve the following specific objectives: significantly reduce the average delay time of the road network, improve the convergence speed and the accuracy of the optimized solution, and enhance the robustness and adaptability of the algorithm to complex, large-scale road network optimization problems, all within a limited and expensive simulation evaluation budget.

[0054] Example 1

[0055] In this embodiment 1, a traffic signal timing system based on surrogate-assisted hierarchical optimization is first provided, including: an acquisition module for acquiring real traffic flow information of the urban road network and phase information of traffic lights; an encoding module for combining each phase timing parameter with the traffic light phase information to encode a complete initial traffic signal timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm; and an iterative search module for performing iterative search using an RBF surrogate model-assisted improved artificial bee colony optimization algorithm; the artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes; during the optimization process, the artificial bee colony optimization algorithm selects based on the prediction results of the surrogate model, retains the best solution, and selects the next best solution based on the filling criterion. The top-ranked schemes are simulated and verified to guide the search towards the optimal region. The simulation module is used to input the optimized timing scheme into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, it simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. The real traffic system is decomposed into two categories: inherent features and dynamic features, and a digital mapping to the SUMO simulation environment is completed. Inherent features refer to the fixed road network topology, which is converted into a SUMO road network file to form the physical basis of the simulation. Dynamic features refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. The two types of features together characterize the spatiotemporal evolution of traffic status, realizing the mapping from the real traffic system to the high-fidelity simulation system.

[0056] The update module is used to take the road network evaluation index output by the SUMO micro-simulation platform as real performance feedback, update the RBF proxy model, and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

[0057] like Figures 1 to 3As shown, in this embodiment, the above-described system is used to implement a traffic signal timing method based on surrogate-assisted hierarchical optimization, including: collecting real traffic flow information of the urban road network and obtaining the phase information of traffic lights; combining the timing parameters of each phase with the phase information of traffic lights to encode a complete initial traffic signal timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm; using the RBF surrogate model to assist the improved artificial bee colony optimization algorithm in iterative search; the artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes; during the optimization process, the artificial bee colony optimization algorithm selects based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes for simulation verification through a filling criterion, guiding the search to converge toward the optimal region; and inputting the optimized timing scheme into the SUMO microscopic simulation platform. Based on the road network topology and traffic flow demand information, the system simulates the operation of vehicles in the actual road network and outputs road network evaluation indicators. The real-world traffic system is decomposed into two categories: inherent characteristics and dynamic characteristics, completing a digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. These two types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real-world traffic system to a high-fidelity simulation system. The road network evaluation indicators output by the SUMO micro-simulation platform are used as real performance feedback to update the RBF proxy model and guide the optimization algorithm to adjust its search strategy, entering the next iteration until the termination condition is met, ultimately outputting the optimal traffic light timing scheme.

[0058] The improved artificial bee colony algorithm assisted by the RBF surrogate model includes the following steps: First, N candidate timing schemes are generated through Latin hypercube sampling. Each candidate timing scheme obtains its precise fitness value through evaluation. These evaluated schemes and their prediction function values ​​constitute the initial database DB. The number of evaluations, NFEs, is initialized to N. The RBF surrogate model is trained using N optimal samples selected from the DB. The improved ABC algorithm is then used to perform the search, sequentially executing the improved mercenary bee phase, determining whether to enter the improved observation bee phase based on probability p, determining whether to trigger the reinitialization of scout bees based on trial constraints, and locating the optimal individual through a filling criterion for real simulation evaluation to update the DB. Next, the local search phase begins. The GP surrogate model is trained using N optimal samples selected from the DB to establish a local optimal region. The DE / current-to-best / 1 operator is used for local search, and the optimal individual is located through the LCB filling criterion for real evaluation and DB update. Finally, it is determined whether the maximum number of function evaluations, MaxFEs, has been reached. If not, the algorithm returns to the global search phase to continue iterating; otherwise, the algorithm terminates and the final result is output.

[0059] The search strategy of the ABC algorithm is improved hierarchically and applied to the hired bee stage. Relying on a novel update mechanism that introduces random variables, the global exploration capability of the population is enhanced, efficiently uncovering untraversed high-quality regions in the solution space and providing a high-quality initial population for subsequent local fine-grained searches. The improved search strategy is as follows: ;in, As the perturbation weight, it is a random number between [-1, 1]; , They are distinct random integers between (0, N); This represents the green light duration for the j-th phase of the i-th test timing scheme; , This represents the green light duration of the j-th phase of a timing scheme for two parent generations randomly selected from the population.

[0060] A local search strategy is introduced into the ABC algorithm to specifically enhance its local optimization performance, enabling rapid convergence to the precise location of the optimal solution and achieving an effective balance between global wide-area search and local fine-grained optimization. The local search strategy is as follows: ;in, This represents the j-th dimension component of the current best individual. As the perturbation weight, it is a random number between [-1, 1]; N is the population size, and in signal timing, the population refers to the number of candidate timing schemes; Let represent the j-th dimension component of the k-th individual randomly selected from the population.

[0061] The specific steps of applying the GP model to the local search phase of the algorithm include: First, selecting N optimal timing schemes with the minimum average delay from the database DB, calculating the range of green light durations for each phase in the population, and adaptively determining the lower bound LB and upper bound UB of the local search for each phase, using the duration of each phase of the current optimal timing scheme as the center, constructing the optimal local region, and focusing the search space on the neighborhood of high-potential timing schemes; Second, selecting N optimal samples from the DB to train a local GP surrogate model, establishing a mapping from timing schemes to average delay, enabling the model to predict the mean delay and uncertainty variance of candidate schemes; Third, generating NP candidate timing schemes within the local optimal region through Latin hypercube sampling as the initial population, and using the trained GP model to quickly predict the delay performance of each scheme, characterizing the fitness distribution of the local solution space; Fourth, assigning differential evolution to each individual in the population. The scaling factor F and crossover probability CR are used to generate experimental timing schemes using the DE / current-to-best / 1 operator. The phase duration of the current best scheme is used as the benchmark to guide the direction of mutation. The step size is controlled by the scaling factor. After binomial crossover, new candidate schemes are generated, and their fitness is evaluated using the GP model. In the fifth step, the GP predicted delay of the experimental scheme is compared with that of the parent scheme, and the high-quality timing scheme with the smaller average delay is retained for the next generation. In the sixth step, the mean predicted delay and uncertainty variance of each candidate scheme are comprehensively considered by the lower confidence bound (LCB) criterion, and high-potential timing schemes with low predicted delay and high uncertainty are selected first. In the seventh step, the optimal potential scheme selected by the LCB criterion is evaluated by SUMO real traffic simulation to obtain the accurate average delay. The timing scheme and its real performance are stored in the database DB, the NFEs count is updated, and the local search iteration is completed.

[0062] In this embodiment, the DE / current-to-best / 1 mutation operator is selected to complete the local population update. This operator uses the best individual in the current population as the mutation reference vector, and regulates the direction of population iteration mutation through the guiding role of the best individual. It can maintain population diversity, avoid premature convergence, and ensure a stable convergence speed of the algorithm, achieving a dynamic balance between exploration and development capabilities. The specific update mechanism is as follows:

[0063]

[0064]

[0065] in, Represents the best individual in the current population. It is a control factor. , Two different individuals randomly selected from the population, Represents the individuals after the crossover. Represents the parent generation; It is a uniformly distributed random number. Let represent the crossover probability in the j-th dimension.

[0066] Example 2

[0067] In this second embodiment, to address the high cost of simulation evaluation in urban road network traffic signal timing optimization, a traffic signal timing method based on surrogate-assisted hierarchical optimization is proposed. For example... Figure 4 As shown, the method specifically includes the following steps:

[0068] Step 1: Obtain traffic flow information: Collect real traffic flow information of the urban road network through traffic detectors (such as loop detectors, video detectors, floating car GPS data, etc.), including basic data such as traffic volume, vehicle speed, and turning at each intersection. At the same time, obtain the phase information of traffic lights (such as the number of phases, phase sequence, and green light duration of each phase) to provide data support for the subsequent timing scheme generation.

[0069] Step 2: Construct a traffic signal timing scheme: Combine the timing parameters of each phase (green light duration, yellow light duration, red light duration, cycle duration, etc.) with the traffic signal phase information, encode them to form a complete traffic signal timing scheme, and use it as the decision variable for the optimization algorithm.

[0070] Step 3: Searching for the optimal timing scheme using an optimization algorithm: The optimization algorithm performs an iterative search. The optimization algorithm is responsible for exploring candidate timing schemes in the solution space, while the surrogate model is responsible for quickly predicting the performance of candidate schemes, replacing time-consuming real simulation evaluation. During the optimization process, the algorithm selects solutions based on the prediction results of the surrogate model, retaining optimal solutions, and selecting a small number of key schemes for real simulation verification through a filling criterion, gradually guiding the search towards the optimal region.

[0071] Step 4: Building the simulation environment and conducting performance evaluation: The real-world traffic system is decomposed into two categories: inherent features and dynamic features, completing the digital mapping to the SUMO simulation environment. Inherent features refer to fixed road network topology (intersection layout, lane division, road segment length, etc.), which are collected through field surveys or OpenStreetMap and converted into SUMO road network files (net.xml), forming the physical foundation of the simulation. Dynamic features refer to real-time changing traffic flow (time-of-day flow, turning ratio, vehicle type composition) and traffic light information (phase sequence, green light ratio, cycle duration), which are collected through detection equipment and converted into SUMO traffic demand files (rou.xml) after cleaning and extraction, serving as the dynamic input for the simulation. These two types of features together characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real-world traffic system to a high-fidelity simulation system.

[0072] The optimized timing scheme is input into the SUMO micro-simulation platform. The simulation system simulates the vehicle's operating status in the actual road network based on the input information such as road network topology and traffic flow demand, and outputs road network evaluation indicators (such as average delay, capacity, number of stops, and travel time).

[0073] Step 5: Feedback Update and Iterative Optimization: The road network evaluation index output by the simulation system is used as real performance feedback to update the surrogate model (expand the training sample library, retrain the RBF / GP model), and guide the optimization algorithm to adjust the search strategy and enter the next round of iteration until the termination condition is met (such as reaching the maximum number of simulation evaluations or performance convergence), and finally output the optimal traffic signal timing scheme.

[0074] Specifically, this method constructs a hierarchical improved surrogate-assisted artificial bee colony (HSAABC) optimization framework. This framework works collaboratively through two core stages: the first stage is an improved ABC algorithm based on the RBF surrogate model for global exploration; the second stage is an adaptive optimal local region differential evolution (DE) search guided by the GP surrogate model for local development. Through information sharing and collaborative iteration, these two stages effectively reduce the number of expensive real-world simulations for evaluation, while simultaneously improving the algorithm's ability to escape local optima and approach the global optimum.

[0075] The pseudocode for the RBF proxy model-assisted improved artificial bee colony algorithm (RBFABC) shown in Table 1 is as follows (the greedy selection algorithm 4 in step 5 of the table represents the greedy selection algorithm in the prior art, and the specific algorithm will not be described again). The formula used in step 3 to update the new individual is specifically the formula in the improved strategy in Example 1. The formula used in step 16 is specifically the one from Example 1. The HSAABC algorithm first generates N candidate timing schemes through Latin hypercube sampling, ensuring coverage in the solution space. Each initial scheme obtains its precise fitness value through expensive real-precise evaluation. These precisely evaluated schemes and their function values ​​constitute the initial database DB. The number of evaluations, NFEs, is initialized to N to track the total number of expensive real-precise evaluations. N optimal samples are selected from the DB to train an RBF surrogate model, which is used to assist the improved ABC algorithm in the search. The improved mercenary bee phase is executed sequentially, followed by a probability p-based determination of whether to enter the improved observer bee phase, a trial-based determination of whether to trigger scout bee reinitialization, and the optimal individual is located and evaluated in real simulation using the filling criterion to update the DB. Then, the local search phase begins, where N optimal samples are selected from the DB to train a GP surrogate model, establishing a local optimal region [LB, UB]. The DE / current-to-best / 1 operator is used for local search, and the optimal individual is located and evaluated in real simulation using the LCB filling criterion to update the DB. Finally, it is determined whether the maximum number of function evaluations, MaxFEs, has been reached. If not, the algorithm returns to the global search phase to continue iterating; otherwise, the algorithm terminates and outputs the final result.

[0076] Table 1

[0077]

[0078] When the traffic light in one direction is green (allowing traffic), the other directions must be red (stopping traffic). The duration of the yellow light and the duration of the all-red light are generally preset fixed values ​​to ensure traffic safety. Therefore, the green light duration of each phase determines the signal cycle of the entire traffic light. Based on this inherent relationship between the phases of the traffic light, this embodiment considers simplifying the encoding of only the green light duration of each phase.

[0079] In practical urban road network traffic signal control optimization projects, the industry generally uses integer vectors to encode signal cycle parameters. The dimension of this integer vector corresponds one-to-one with the number of signal phases at the intersection, and the value of each integer element within the vector represents the passage duration of the corresponding signal phase, uniformly measured in seconds. This encoding method aligns with the hardware operating logic of real traffic management scenarios. Since the minimum timing and control unit of existing traffic signal controllers is an integer second, the passage duration for each phase only supports integer seconds and cannot be set to decimal seconds. Therefore, using integer vector encoding can accurately match the requirements of actual engineering applications, ensuring the feasibility of the timing scheme output by the optimization algorithm.

[0080] For single-objective traffic signal optimization models, average vehicle delay time is the core evaluation indicator for assessing the operational efficiency of the traffic network. It can intuitively represent the overall smoothness of traffic flow, and its value can directly and objectively reflect the level of road traffic congestion and the quality of public travel experience. To ensure the engineering feasibility of the traffic signal control scheme, and to effectively simplify the optimization problem and reduce the complexity of the algorithm, this embodiment only performs precise optimization on the green light duration of each phase at the intersection. Other signal control parameters, such as phase sorting method, total signal cycle duration, and network phase difference, are all fixed values ​​and not iteratively optimized. The constraints and solution process of the above optimization variables can be defined and represented by the following formula: .

[0081] In solving for the optimal timing scheme of the objective function, it is necessary to apply reasonable constraints to the optimization variables based on the actual operating rules and safety regulations of real road traffic scenarios. The value of the green light duration directly affects the safety and overall traffic efficiency of the intersection, and there is a clear reasonable range of values. If the green light duration is set too short, pedestrians and non-motorized vehicles cannot safely cross the intersection within the effective time, which can easily lead to intersection conflicts and traffic safety hazards; if the green light duration is set too long, it will cause a significant increase in the waiting time of traffic in all directions, resulting in ineffective congestion and waste of road network resources, and significantly reducing the overall traffic throughput efficiency of the intersection. Based on the above actual engineering constraints, the following constraints are set for the variable values ​​of the optimization model:

[0082]

[0083]

[0084]

[0085] in, Represents a vector of decision variables (an integer vector consisting of the green light duration for each phase); Includes the duration of the yellow light and the duration of the all-red light; C is the cycle duration, and g is the green light duration. C min C represents the minimum period duration. max This represents the maximum period duration. g min G represents the minimum period duration. max This indicates the maximum duration of the period.

[0086] To accurately quantify and evaluate the practical application effect of single-objective traffic signal control schemes, this embodiment uses the open-source microscopic traffic simulation tool SUMO to build a simulation test model. SUMO possesses high-precision microscopic simulation capabilities, enabling second-by-second fine-grained modeling of vehicle behavior, and providing a repeatable and flexibly adjustable simulation experimental environment for traffic signal optimization research. This simulation tool can quickly load various signal timing schemes and embed them into the road network model, outputting core traffic evaluation indicators such as vehicle delays and traffic flow in real time.

[0087] In the optimization framework of this embodiment, SUMO serves as a black-box evaluation unit in an iterative loop to complete the scheme performance evaluation. Each time the optimizer generates a new set of timing decision variables, the system automatically calls SUMO to complete the full-process traffic simulation and outputs the corresponding road network performance indicators, which serve as the basis for determining the objective function. The specific expression is as follows: Where AD is the average delay of vehicles on the road network output by the simulation, which is the core evaluation index; p is the vector of endogenous variables in the simulation; and q is the vector of exogenous fixed variables.

[0088] The standard ABC algorithm possesses strong global exploration capabilities, but its local exploration capabilities are significantly weak. Traditional ABC algorithms employ only a single search strategy for population updates, with the search step size determined by the weighted difference between the current solution and randomly selected solutions from the population. Since the quality of randomly selected individuals cannot be predetermined, and the weight factor ϕ is randomly generated within the range [-1,1], the traditional search mechanism exhibits excessive randomness. While this characteristic broadens the global search scope, it significantly reduces the algorithm's ability to perform precise local searches, easily leading to insufficient convergence accuracy of the optimal solution. In this embodiment, the proposed RBF surrogate model assists in improving the artificial bee colony algorithm (RBFABC), integrating the radial basis function (RBF) surrogate model with the artificial bee colony (ABC) algorithm to intelligently upgrade the traditional optimization iterative process, balancing search efficiency and solution accuracy. This algorithm uses the ABC algorithm to complete the global search because it has the advantages of fewer parameters, strong robustness, and good coverage of swarm intelligence search, and can efficiently traverse the complete decision space. At the same time, the RBF surrogate model has simple computational logic, stable prediction accuracy, and excellent adaptability to complex optimization problems with high dimension and strong nonlinearity. It can accurately replace high-cost real simulation evaluation and is an ideal surrogate model for this optimization scenario.

[0089] To balance the global exploration and local exploitation capabilities of the algorithm and improve overall optimization performance, this embodiment improves the search strategy of the ABC algorithm in a layered manner. The improved search strategy is applied to the hired bee stage, relying on a novel update mechanism that introduces random variables to enhance the global exploration capability of the population, efficiently discovering high-quality regions not yet explored in the solution space, and providing a high-quality initial population for subsequent local fine-grained searches. The improved search strategy is as follows: ;in, As the perturbation weight, it is a random number between [-1, 1]; , It is a distinct random integer between (0, N]. This represents the green light duration for the j-th phase of the i-th test timing scheme; , This represents the green light duration of the j-th phase of a timing scheme for two parent generations randomly selected from the population.

[0090] Simultaneously, a local search strategy is introduced to specifically enhance the algorithm's local optimization performance, enabling rapid convergence to the precise location of the optimal solution and achieving an effective balance between global wide-area search and local fine-grained optimization. The local search strategy is as follows: ;in, This represents the j-th dimension component of the current best individual. As the perturbation weight, it is a random number between [-1, 1]; N is the population rule, and in signal timing, the population refers to the number of candidate timing schemes. Let represent the j-th dimension component of the k-th individual randomly selected from the population.

[0091] In summary, this embodiment introduces an RBF surrogate model to pre-screen offspring individuals in the traditional artificial bee colony algorithm's hired bee and observation bee stages, thereby reducing the computational overhead of high-precision simulation evaluation. The algorithm relies on the trained RBF model to quickly predict the performance of offspring individuals, preemptively eliminating inferior individuals with poor optimization potential and performance, significantly reducing the number of meaningless high-precision real-world evaluations, substantially decreasing overall iteration computation time, and effectively improving the overall efficiency of the algorithm.

[0092] The algorithm uses a dedicated database (DB) to uniformly store the training samples of the RBF model. To balance modeling efficiency and prediction accuracy, the maximum number of samples that can be stored in the database is limited to N. The core function of the RBF surrogate model is to accurately fit the objective function distribution of the population neighborhood in a low-cost manner, replacing high-cost simulation evaluation. Therefore, the training samples must closely match the high-quality solution region sought by the algorithm's iterative optimization. Based on this principle, this embodiment consistently selects the N groups of individuals with the best performance from the database DB to construct the training set, ensuring the surrogate model's accuracy in predicting the neighborhood of the optimal solution.

[0093] During the mercenary bee iteration phase, each parent individual generates corresponding offspring individuals. Subsequently, the trained RBF model is used to quickly predict the objective function values ​​of all offspring. For ordinary parent individuals that are not globally optimal, if the offspring's prediction performance is better than the parent's, the iteration update of parent and offspring individuals is directly completed, quickly achieving the survival of the fittest in the population.

[0094] For the current best individual in the population, this embodiment employs a differentiated update strategy to avoid optimization bias caused by prediction errors in the surrogate model. For offspring generated based on the globally optimal individual, the RBF prediction results are no longer relied upon; instead, a high-precision true objective function is directly used to complete the performance evaluation, and the sample is archived in the database DB. The globally optimal solution is only updated when the actual evaluation performance of the offspring is better than that of the parent's best individual, effectively avoiding false optimal solutions caused by model prediction bias and ensuring the accuracy and stability of the algorithm's iterative optimization.

[0095] In the subsequent observation bee phase, the RBF pre-screening mechanism is reused to quickly select individuals from the neighborhood search. The replacement rules for parent and offspring individuals remain consistent with those in the hired bee phase, and the entire process relies on the proxy model to reduce unnecessary computation. Simultaneously, the RBF model is continuously retrained using the optimal samples during algorithm iteration to dynamically improve the model's prediction accuracy and adapt to the optimization changes during population iteration.

[0096] Performing true function evaluation on certain promising individuals can actually enhance the reliability of the surrogate model. After individual updates, individuals with the smallest RBF model prediction function value are selected, evaluated using the expensive exact function, and then archived into the database. The selection strategy is as follows: ;in, Individuals who conduct real assessments, Describe the objective function Approximate value.

[0097] Computational resources are typically limited in urban road network signal optimization scenarios, making real simulation evaluation sample data extremely valuable. The limited number of evaluations directly determines the final optimization effect. Therefore, this invention retains all individual samples that have undergone real, high-precision evaluations, continuously enriching the high-quality sample library. This allows the surrogate model to maintain optimal fitting accuracy under limited computational resources, thereby achieving optimal iterative optimization of the signal timing scheme within a controllable computational cost.

[0098] While the aforementioned global search mechanism based on the RBF surrogate model exhibits good convergence performance, its global exploration capability is limited. It is prone to premature convergence during algorithm iteration, struggling to escape local optima and failing to fully explore superior global optimal timing schemes within the solution space. To address this deficiency, this invention introduces a Gaussian process (GP) model as a local surrogate model. Compared to a single-prediction surrogate model, the GP model possesses unique technical advantages, not only providing approximate predictions of individual fitness but also outputting quantified variance information of the corresponding prediction results. By integrating and utilizing both the prediction mean and variance information, the algorithm's global exploration capability can be effectively enhanced, helping it break free from local optima constraints, broaden the search coverage, and further improve global optimization accuracy.

[0099] Based on the above advantages, this embodiment applies the GP model to the local search stage of the algorithm. The specific pseudocode implementation process is shown in Table 2. In the model construction stage, the system selects N sets of high-quality optimal samples from the sample database DB to complete the training and construction of the local GP proxy model. Simultaneously, it uses Latin hypercube sampling (LHS) to generate an initial population of NP size, and relies on the trained local GP model to perform a pre-evaluation of population performance, accurately characterizing the fitness distribution features of the current local solution space, providing data support for subsequent refined local search. As shown in Table 2, the specific steps for applying the GP model to the local search stage of the algorithm include: First, selecting N optimal timing schemes with the minimum average delay from the database DB, calculating the range of green light duration for each phase in the population, and adaptively determining the lower bound LB and upper bound UB of the local search for each phase with the duration of each phase of the current optimal timing scheme as the center, constructing the optimal local region, and focusing the search space on the neighborhood of high-potential timing schemes; Second, selecting N optimal samples from the DB to train the local GP surrogate model, establishing a mapping from timing schemes to average delay, enabling the model to predict the mean delay and uncertainty variance of candidate schemes; Third, using Latin hypercube sampling in [LB, Within the range of UB, NP candidate timing schemes are generated as the initial population, and the trained GP model is used to quickly predict the delay performance of each scheme and characterize the fitness distribution of the local solution space; in the fourth step, a differential evolution scaling factor F and crossover probability CR are assigned to each individual in the population, and the DE / current-to-best / 1 operator is used to generate experimental timing schemes. The mutation direction is guided by the phase duration of the current best scheme, and the step size is controlled by the scaling factor. After binomial crossover, new candidate schemes are generated, and their fitness is evaluated by the GP model; in the fifth step... The first step involves comparing the GP predicted delay of the experimental scheme with that of the parent scheme, retaining the superior timing scheme with the smaller average delay for the next generation; the second step involves comprehensively considering the mean predicted delay and uncertainty variance of each candidate scheme using the lower confidence bound (LCB) criterion, prioritizing the selection of high-potential timing schemes with low predicted delay and high uncertainty; the third step involves performing SUMO real traffic simulation evaluation on the optimal potential scheme selected by the LCB criterion to obtain the accurate average delay, storing the timing scheme and its actual performance in the database DB, updating the NFEs count, and completing this round of local search iteration.

[0100] Table 2

[0101]

[0102] Traditional intelligent optimization algorithms generally suffer from redundancy in the search interval and excessively broad exploration range during global optimization. Algorithm iterations require traversing the entire solution space, resulting in numerous invalid iterations and redundant calculations. This is particularly problematic in high-dimensional scenarios like large-scale urban road network signal optimization, significantly increasing simulation computational pressure and reducing algorithm convergence efficiency. To address this deficiency, this invention innovatively designs an adaptive local region determination mechanism. Based on population iteration data, it dynamically identifies and locks onto core local regions with high optimization potential within the solution space, achieving adaptive shrinkage and precise limitation of the search space.

[0103] To further focus on the effective search area and reduce the overhead of ineffective iterations, this embodiment designs an optimal local region determination mechanism to accurately lock the local core region with high optimization potential in the solution space, realize adaptive compression of the search space, and enable the algorithm to focus on iterative optimization of the interval where the global optimal solution exists with a high probability, thereby reducing the computational resource consumption caused by ineffective exploration.

[0104] The solution space local region identification method is shown below, which is used to define the value boundaries of individuals in each dimension.

[0105]

[0106] Among them, variables For the d-th dimension variable corresponding to the i-th individual, and represents the upper and lower boundaries of the d-th dimension of the search space; N represents the population size.

[0107] The constraints of the optimal local region are defined as follows:

[0108]

[0109] In the formula, L represents the upper bound, U represents the lower bound, D is the global dimension, d is the index of the remaining dimension after removing the j-th dimension, and LB and UB are the lower and upper bounds of the optimal local space, respectively.

[0110] Differential Evolution (DE) algorithm is simple in structure and highly efficient in computation, demonstrating excellent optimization performance in various benchmark functions and engineering optimization scenarios. In this embodiment, the DE / current-to-best / 1 mutation operator is used to complete the local population update. This operator uses the best individual in the current population as the mutation reference vector, and regulates the direction of population iteration mutation through the guiding role of the best individual. It can maintain population diversity, avoid premature convergence, and ensure a stable convergence speed of the algorithm, achieving a dynamic balance between exploration and development capabilities. The specific update mechanism is shown below.

[0111]

[0112]

[0113] in, Represents the best individual in the current population. It is a control factor. , Two different individuals randomly selected from the population, Represents the individuals after the crossover. Represents the parent generation. It is a uniformly distributed random number. Let represent the crossover probability in the j-th dimension. In the formula, each parameter corresponds to the current optimal individual in the population and the iterative control coefficient, jointly regulating the population mutation and update process.

[0114] The core principle of this operator is to use the best individual in the current iterative population as the reference vector for mutation, and to guide the overall iterative direction of the population based on the high-quality solution characteristics of the best individual, thereby replacing the traditional random mutation mode and effectively avoiding the iterative disorder problem caused by blind search.

[0115] Traditional surrogate model screening methods rely solely on a single predicted mean to determine individual merit, failing to quantify the uncertainty of prediction results and easily overlooking high-potential individuals. Furthermore, they are susceptible to model prediction biases, leading to optimization failures. This invention introduces a lower confidence bound criterion to construct an individual pre-screening mechanism. This criterion comprehensively considers individual prediction performance and uncertainty characteristics, prioritizing the screening of potential individuals for real-world simulation evaluation. This improves local search accuracy while further reducing the number of expensive simulation calls, balancing algorithm optimization accuracy and computational efficiency.

[0116] By combining the mean and variance information of the GP model, high-potential individuals can be accurately screened. The specific calculation method is shown in the formula. ,in, This represents the lower confidence boundary criterion. This represents the mean prediction of the GP model. This represents the relevant prediction variance, where h is a constant.

[0117] The best potential individuals selected through this criterion will undergo a real high-precision objective function evaluation and be archived in the database DB. The sample library will be continuously updated, the accuracy of the surrogate model will be optimized, and the overall optimization performance and engineering adaptability of the algorithm will be further improved.

[0118] Example 3

[0119] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the traffic signal timing method based on agent-assisted hierarchical optimization as described above. The method includes:

[0120] Collect real traffic flow information from the urban road network and obtain phase information from traffic lights;

[0121] The timing parameters of each phase are combined with the traffic light phase information and encoded to form a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm.

[0122] An improved artificial bee colony optimization algorithm is used to perform iterative search using an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes through the filling criterion for simulation verification, guiding the search to converge toward the optimal region.

[0123] The optimized timing scheme is input into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, the system simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. The real traffic system is decomposed into two categories: inherent characteristics and dynamic characteristics, completing a digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. Both types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to a high-fidelity simulation system.

[0124] The road network evaluation index output by the SUMO micro-simulation platform is used as real performance feedback to update the RBF proxy model and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

[0125] Example 4

[0126] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, and the memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the traffic signal timing method based on agent-assisted hierarchical optimization as described above, the method including:

[0127] Collect real traffic flow information from the urban road network and obtain phase information from traffic lights;

[0128] The timing parameters of each phase are combined with the traffic light phase information and encoded to form a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm.

[0129] An improved artificial bee colony optimization algorithm is used to perform iterative search using an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes through the filling criterion for simulation verification, guiding the search to converge toward the optimal region.

[0130] The optimized timing scheme is input into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, the system simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. The real traffic system is decomposed into two categories: inherent characteristics and dynamic characteristics, completing a digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. Both types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to a high-fidelity simulation system.

[0131] The road network evaluation index output by the SUMO micro-simulation platform is used as real performance feedback to update the RBF proxy model and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

[0132] Example 5

[0133] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the traffic signal timing method based on agent-assisted hierarchical optimization as described above. The method includes:

[0134] Collect real traffic flow information from the urban road network and obtain phase information from traffic lights;

[0135] The timing parameters of each phase are combined with the traffic light phase information and encoded to form a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm.

[0136] An improved artificial bee colony optimization algorithm is used to perform iterative search using an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes through the filling criterion for simulation verification, guiding the search to converge toward the optimal region.

[0137] The optimized timing scheme is input into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, the system simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. The real traffic system is decomposed into two categories: inherent characteristics and dynamic characteristics, completing a digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. Both types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to a high-fidelity simulation system.

[0138] The road network evaluation index output by the SUMO micro-simulation platform is used as real performance feedback to update the RBF proxy model and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

[0139] In summary, the traffic signal timing method based on surrogate-assisted hierarchical optimization described in this invention proposes a traffic signal timing optimization framework based on surrogate-assisted hierarchical optimization. This framework decomposes the optimization process into two tightly coupled layers: global exploration and local exploitation, significantly improving optimization efficiency and accuracy under limited and expensive evaluation budgets. In the global exploration phase, an improved artificial bee colony algorithm assisted by an RBF surrogate model is employed, including: using the RBF model to quickly pre-screen candidate solutions to reduce simulation costs; introducing two improved ABC search strategies that respectively emphasize exploration and exploitation; and utilizing the most valuable real evaluation resources to correct the model and update the optimal solution based on the filling criterion and greedy selection strategy. In the local exploitation phase, an adaptive optimal local region differential evolution search method guided by a Gaussian process surrogate model is adopted. This includes: adaptively determining the optimal local region based on the population state to narrow the search range; leveraging the unique advantage of the GP model in providing predictive uncertainty, intelligently balancing local exploration and exploitation through the lower confidence bound (LCB) criterion; and using the DE algorithm as a local searcher to achieve rapid and refined approximation of the global optimal solution.

[0140] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0143] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0144] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A traffic signal timing method based on agent-assisted hierarchical optimization, characterized in that, include: Collect real traffic flow information from the urban road network and obtain phase information from traffic lights; The timing parameters of each phase are combined with the traffic light phase information and encoded to form a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm. An improved artificial bee colony optimization algorithm is used to perform iterative search using an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes through the filling criterion for simulation verification, guiding the search to converge toward the optimal region. The optimized timing scheme is input into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, the system simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. The real traffic system is decomposed into two categories: inherent characteristics and dynamic characteristics, completing a digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. The two types of features together characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to the high-fidelity simulation system; The road network evaluation index output by the SUMO micro-simulation platform is used as real performance feedback to update the RBF proxy model and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

2. The traffic signal timing method based on agent-assisted hierarchical optimization according to claim 1, characterized in that, The improved artificial bee colony algorithm assisted by the RBF surrogate model includes the following steps: First, N candidate timing schemes are generated through Latin hypercube sampling. Each candidate timing scheme obtains its precise fitness value through evaluation. These evaluated schemes and their prediction function values ​​constitute the initial database DB. The number of evaluations, NFEs, is initialized to N. The RBF surrogate model is trained using N optimal samples selected from the DB. The improved ABC algorithm is then used to perform the search, sequentially executing the improved mercenary bee phase, determining whether to enter the improved observation bee phase based on probability p, determining whether to trigger the reinitialization of scout bees based on trial constraints, and locating the optimal individual through a filling criterion for real simulation evaluation to update the DB. Next, the local search phase begins. The GP surrogate model is trained using N optimal samples selected from the DB to establish a local optimal region. The DE / current-to-best / 1 operator is used for local search, and the optimal individual is located through the LCB filling criterion for real evaluation and DB update. Finally, it is determined whether the maximum number of function evaluations, MaxFEs, has been reached. If not, the algorithm returns to the global search phase to continue iterating; otherwise, the algorithm terminates and the final result is output.

3. The traffic signal timing method based on agent-assisted hierarchical optimization according to claim 2, characterized in that, The search strategy of the ABC algorithm is improved hierarchically and applied to the hired bee stage. Relying on a novel update mechanism that introduces random variables, the global exploration capability of the population is enhanced, efficiently uncovering untraversed high-quality regions in the solution space and providing a high-quality initial population for subsequent local fine-grained searches. The improved search strategy is as follows: ;in, As the perturbation weight, it is a random number between [-1, 1]; , They are distinct random integers between (0, N); This represents the green light duration for the j-th phase of the i-th test timing scheme; , This represents the green light duration of the j-th phase of a timing scheme for two parent generations randomly selected from the population.

4. The traffic signal timing method based on agent-assisted hierarchical optimization according to claim 3, characterized in that, A local search strategy is introduced into the ABC algorithm to specifically enhance its local optimization performance, enabling rapid convergence to the precise location of the optimal solution and achieving an effective balance between global wide-area search and local fine-grained optimization. The local search strategy is as follows: ;in, This represents the j-th dimension component of the current best individual. As the perturbation weight, it is a random number between [-1, 1]; N is the population size, and in signal timing, the population refers to the number of candidate timing schemes; Let represent the j-th dimension component of the k-th individual randomly selected from the population.

5. The traffic signal timing method based on agent-assisted hierarchical optimization according to claim 2, characterized in that, The specific steps of applying the GP model to the local search phase of the algorithm include: First, selecting N optimal timing schemes with the minimum average delay from the database DB, calculating the range of green light durations for each phase in the population, and adaptively determining the lower bound LB and upper bound UB of the local search for each phase, using the duration of each phase of the current optimal timing scheme as the center, constructing the optimal local region, and focusing the search space on the neighborhood of high-potential timing schemes; Second, selecting N optimal samples from the DB to train a local GP surrogate model, establishing a mapping from timing schemes to average delay, enabling the model to predict the mean delay and uncertainty variance of candidate schemes; Third, generating NP candidate timing schemes within the local optimal region through Latin hypercube sampling as the initial population, and using the trained GP model to quickly predict the delay performance of each scheme, characterizing the fitness distribution of the local solution space; Fourth, assigning differential evolution to each individual in the population. The scaling factor F and crossover probability CR are used to generate experimental timing schemes using the DE / current-to-best / 1 operator. The phase duration of the current best scheme is used as the benchmark to guide the direction of mutation. The step size is controlled by the scaling factor. After binomial crossover, new candidate schemes are generated, and their fitness is evaluated using the GP model. In the fifth step, the GP predicted delay of the experimental scheme is compared with that of the parent scheme, and the high-quality timing scheme with the smaller average delay is retained for the next generation. In the sixth step, the mean predicted delay and uncertainty variance of each candidate scheme are comprehensively considered by the lower confidence bound (LCB) criterion, and high-potential timing schemes with low predicted delay and high uncertainty are selected first. In the seventh step, the optimal potential scheme selected by the LCB criterion is evaluated by SUMO real traffic simulation to obtain the accurate average delay. The timing scheme and its real performance are stored in the database DB, the NFEs count is updated, and the local search iteration is completed.

6. The traffic signal timing method based on agent-assisted hierarchical optimization according to claim 5, characterized in that, The DE / current-to-best / 1 mutation operator is selected to complete the local population update. This operator uses the best individual in the current population as the mutation reference vector. By guiding the population's iterative mutation direction through the guiding role of the best individual, it can maintain population diversity, avoid premature convergence, and ensure a stable convergence speed of the algorithm, achieving a dynamic balance between exploration and development capabilities. The specific update mechanism is as follows: in, Represents the best individual in the current population. It is a control factor. , Two different individuals randomly selected from the population, Represents the individuals after the crossover. Represents the parent generation; It is a uniformly distributed random number. Let represent the crossover probability in the j-th dimension.

7. A traffic signal timing system based on agent-assisted hierarchical optimization, characterized in that, include: The acquisition module is used to acquire real traffic flow information and traffic light phase information of the urban road network; The encoding module is used to combine the timing parameters of each phase with the traffic light phase information to encode a complete initial traffic light timing scheme, which serves as the decision variable for the artificial bee colony optimization algorithm. The iterative search module is used to perform iterative search using an improved artificial bee colony optimization algorithm assisted by an RBF surrogate model. The artificial bee colony optimization algorithm explores candidate timing schemes in the solution space, and the RBF surrogate model predicts the performance of the candidate schemes. During the optimization process, the artificial bee colony optimization algorithm selects the best solution based on the prediction results of the surrogate model, retains the best solution, and selects the top-ranked schemes for simulation verification through the filling criterion, guiding the search to converge toward the optimal region. The simulation module is used to input the optimized timing scheme into the SUMO micro-simulation platform. Based on the road network topology and traffic flow demand information, it simulates the vehicle operation status in the actual road network and outputs road network evaluation indicators. Specifically, it decomposes the real traffic system into two categories: inherent characteristics and dynamic characteristics, completing the digital mapping to the SUMO simulation environment. Inherent characteristics refer to the fixed road network topology, which is converted into a SUMO road network file, forming the physical basis of the simulation. Dynamic characteristics refer to the real-time changing traffic flow and traffic light information, which is converted into a SUMO traffic demand file as the dynamic input for the simulation. Both types of characteristics jointly characterize the spatiotemporal evolution of traffic conditions, realizing the mapping from the real traffic system to a high-fidelity simulation system. The update module is used to take the road network evaluation index output by the SUMO micro-simulation platform as real performance feedback, update the RBF proxy model, and guide the optimization algorithm to adjust the search strategy, enter the next round of iteration, until the termination condition is met, and finally output the optimal traffic signal timing scheme.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the traffic signal timing method based on agent-assisted hierarchical optimization as described in any one of claims 1-6.

9. A computer device, characterized in that, The system includes a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions executable by the processor, and the processor calling the program instructions to execute the traffic signal timing method based on agent-assisted hierarchical optimization as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the traffic signal timing method based on agent-assisted hierarchical optimization as described in any one of claims 1-6.

Images