Driving decision model optimization method and device, computer equipment and storage medium

Abstract

The invention relates to a driving decision model optimization method and device, computer equipment and a storage medium. The method comprises the following steps: determining an initial optimal group corresponding to an initial candidate solution group according to a preset classifier model; determining a segmentation scaling factor by adopting a preset coefficient and an iteration parameter, and improving the evolutionary algorithm by adopting the segmentation scaling factor to obtain an improved evolutionary algorithm; optimizing the initial candidate solution group based on an improved evolutionary algorithm to obtain an optimized candidate solution group, and determining an evolutionary optimal group corresponding to the optimized candidate solution group based on a classifier model; and determining an optimized classifier according to the evolution optimal group and the initial optimal group, and optimizing the driving decision model according to the optimized classifier. By adopting the method, the parameters of the classifier can be optimized, the performance of the classifier is improved, and the driving decision-making model of the driving decision-making accuracy is further improved.

Description

Technical Field

[0001] The present application relates to the field of unmanned driving technology, and in particular to a driving decision model optimization method, device, computer equipment and storage medium. Background Art

[0002] With the development of artificial intelligence technology, driverless technology has also made significant progress. The driverless system is a control system that integrates environmental perception, driving decision-making and adaptive control technologies. Among them, driving decision-making is the "intelligent core" of driverless cars and plays a vital role in ensuring the driving safety and rationality of driverless cars. Therefore, improving the intelligence level of driving decision-making is the top priority in the field of driverless technology.

[0003] The operation of driving decisions depends on the comprehensive collection, accurate integration and in-depth analysis of the surrounding environment information. In essence, driving decisions are a complex classification problem, which requires the system to accurately distinguish and respond to different driving scenarios, and then make correct driving decisions. In order to simplify the complexity of driving decision problems, driving behaviors are classified during the driving decision process, and then the complex driving decision process is transformed into a clear binary classification problem, which enables driverless cars to accurately identify different driving scenarios and make corresponding driving decisions accordingly. Therefore, in the classification process of driving decisions, classifiers become key components.

[0004] However, in practical applications, traditional classifiers are affected to a certain extent by parameter selection. When the classifier parameters are not selected reasonably, the classification performance of the classifier will be greatly reduced, thereby affecting the accuracy of driving decisions. Summary of the invention

[0005] Based on this, it is necessary to provide a driving decision model optimization method, device, computer equipment, computer-readable storage medium and computer program product that can optimize classifier parameters, improve classifier performance, and thus improve the accuracy of driving decisions in response to the above technical problems.

[0006] In a first aspect, the present application provides a driving decision model optimization method, comprising:

[0007] According to the preset classifier model, determine the initial optimal group corresponding to the initial candidate solution group;

[0008] The segmented scaling factor is determined by using a preset coefficient and an iterative parameter, and the evolutionary algorithm is improved by using the segmented scaling factor to obtain an improved evolutionary algorithm;

[0009] The initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain the optimized candidate solution group, and the evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model;

[0010] An optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, and the driving decision model is optimized according to the optimized classifier.

[0011] In one embodiment, determining the segmented scaling factor using the preset coefficient and the iteration parameter includes:

[0012] The evolutionary algorithm is processed in stages to obtain a multi-stage evolutionary algorithm;

[0013] Determine the preset coefficients corresponding to the evolutionary algorithm at each stage, and determine the segmented scaling factors corresponding to the evolutionary algorithm at each stage according to the preset coefficients and iteration parameters corresponding to the evolutionary algorithm at each stage.

[0014] In one embodiment, determining an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimizing the driving decision model according to the optimized classifier includes:

[0015] Obtain the effect value corresponding to the evolutionary optimal group and the effect value corresponding to the initial optimal group, and compare the effect value corresponding to the evolutionary optimal group with the effect value corresponding to the initial optimal group to obtain the target optimal group;

[0016] The classifier model is adjusted using the target optimal group to obtain an optimized classifier;

[0017] The driving decision model is optimized based on the optimized classifier.

[0018] In one embodiment, the initial candidate solution group is optimized based on the improved evolutionary algorithm, and the optimized candidate solution group includes:

[0019] Performing a perturbation operation on the initial candidate solution group according to the piecewise scaling factor to obtain a first improved candidate solution group;

[0020] The first improved candidate solution group and the initial candidate solution group are reorganized to obtain a second improved candidate solution group.

[0021] In one embodiment, the method further comprises:

[0022] Determine the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group;

[0023] According to the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group, the second improved candidate solution group and the initial candidate solution group are screened to obtain an optimized candidate solution group.

[0024] In one embodiment, before determining the initial optimal group corresponding to the initial candidate solution group according to the preset classifier model, the method further includes:

[0025] Obtain the number of candidate solution groups; generate dimension parameters for each candidate solution in the candidate solution group according to the number of candidate solution groups and a generation algorithm;

[0026] An initial candidate solution group is generated according to the dimension parameters of each candidate solution; wherein each individual in the initial candidate solution group is a key parameter of the classifier model.

[0027] In a second aspect, the present application also provides a driving decision model optimization device, comprising:

[0028] A first determination module is used to determine an initial optimal group corresponding to an initial candidate solution group according to a preset classifier model;

[0029] An improvement module is used to determine a segmented scaling factor using a preset coefficient and an iteration parameter, and to improve the evolutionary algorithm using the segmented scaling factor to obtain an improved evolutionary algorithm;

[0030] The second determination module is used to optimize the initial candidate solution group based on the improved evolutionary algorithm to obtain an optimized candidate solution group, and determine the evolutionary optimal group corresponding to the optimized candidate solution group based on the classifier model;

[0031] The optimization module is used to determine an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimize the driving decision model according to the optimized classifier.

[0032] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0033] According to the preset classifier model, determine the initial optimal group corresponding to the initial candidate solution group;

[0034] The segmented scaling factor is determined by using a preset coefficient and an iterative parameter, and the evolutionary algorithm is improved by using the segmented scaling factor to obtain an improved evolutionary algorithm;

[0035] The initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain the optimized candidate solution group, and the evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model;

[0036] An optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, and the driving decision model is optimized according to the optimized classifier.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0038] According to the preset classifier model, determine the initial optimal group corresponding to the initial candidate solution group;

[0039] The segmented scaling factor is determined by using a preset coefficient and an iterative parameter, and the evolutionary algorithm is improved by using the segmented scaling factor to obtain an improved evolutionary algorithm;

[0040] The initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain the optimized candidate solution group, and the evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model;

[0041] An optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, and the driving decision model is optimized according to the optimized classifier.

[0042] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0043] According to the preset classifier model, determine the initial optimal group corresponding to the initial candidate solution group;

[0044] The segmented scaling factor is determined by using a preset coefficient and an iterative parameter, and the evolutionary algorithm is improved by using the segmented scaling factor to obtain an improved evolutionary algorithm;

[0045] The initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain the optimized candidate solution group, and the evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model;

[0046] An optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, and the driving decision model is optimized according to the optimized classifier.

[0047] The above-mentioned driving decision model optimization method, device, computer equipment, storage medium and computer program product determine the initial optimal group corresponding to the initial candidate solution group according to the preset classifier model; use the preset coefficient and iteration parameter to determine the segmented scaling factor, and use the segmented scaling factor to improve the evolutionary algorithm to obtain the improved evolutionary algorithm; optimize the initial candidate solution group based on the improved evolutionary algorithm to obtain the optimized candidate solution group, and determine the evolutionary optimal group corresponding to the optimized candidate solution group based on the classifier model; determine the optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimize the driving decision model according to the optimized classifier. The driving decision model optimization method can optimize the parameters of the classifier, improve the classifier performance, and thus improve the driving decision accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] FIG1 is an application environment diagram of a driving decision model optimization method according to an embodiment;

[0050] FIG2 is a schematic diagram of a flow chart of a driving decision model optimization method according to an embodiment;

[0051] FIG3 is a schematic diagram of a flow chart of steps for optimizing a driving decision model in one embodiment;

[0052] FIG4 is a structural block diagram of a driving decision model optimization device in one embodiment;

[0053] FIG. 5 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] The driving decision model optimization method provided in the embodiment of the present application can be applied in the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 sends a driving decision model optimization request to the server 104, and the server 104 receives the driving decision model optimization request, and determines the initial optimal group corresponding to the initial candidate solution group according to the preset classifier model; the segmented scaling factor is determined by using the preset coefficient and the iteration parameter, and the evolutionary algorithm is improved by using the segmented scaling factor to obtain the improved evolutionary algorithm; the initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain the optimized candidate solution group, and the evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model; the optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, and the driving decision model is optimized according to the optimized classifier. Among them, the terminal 102 can be, but is not limited to, various vehicles such as cars, motorcycles, and aircraft. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0056] In an exemplary embodiment, as shown in FIG2 , a driving decision model optimization method is provided, which is described by taking the method applied to the server in FIG1 as an example, and includes the following steps 202 to 208 . Among them:

[0057] Step 202: Determine the initial optimal group corresponding to the initial candidate solution group according to the preset classifier model.

[0058] The classifier model is used to classify the input data into different categories. The initial candidate solution group refers to a set of candidate solutions generated at the beginning of the evolutionary algorithm, and each set of candidate solutions is composed of key parameters in the classifier model.

[0059] Optionally, the classifier model can be a support vector machine (SVM) model, or a random forest model, a decision tree model, etc. Exemplarily, the key parameters corresponding to the classifier model are determined, and an initial candidate solution group corresponding to the key parameters is generated. For each candidate solution in the initial candidate solution group, a fitness function is used to evaluate the effect value corresponding to the group of candidate solutions. Multiple effect values ​​corresponding to the initial candidate solution group are obtained, and the multiple effect values ​​corresponding to the initial candidate solution group are compared to determine the highest effect value among the multiple effect values ​​corresponding to the initial candidate solution group, and the candidate solution corresponding to the highest effect value is used as the initial optimal group. Among them, the fitness function is used to measure the classification performance of the classifier model under given parameters. In the SVM classifier, the fitness function can be indicators such as classification accuracy, F1 score or cross-validation error.

[0060] Step 204: Use the preset coefficient and the iteration parameter to determine the segmented scaling factor, and use the segmented scaling factor to improve the evolutionary algorithm to obtain an improved evolutionary algorithm.

[0061] The iteration parameters may include the maximum iteration parameter and the current iteration parameter, and the scaling factor refers to the parameter used to control the degree of individual variation in the evolutionary algorithm. The segmented scaling factor refers to a specific scaling factor value corresponding to each stage determined according to each stage of the evolutionary algorithm after the evolutionary algorithm is divided into multiple stages, that is, the evolutionary algorithm is allowed to dynamically adjust the scaling factor according to the search progress at different stages, and then adjust the degree of variation to adapt to the different optimization requirements of each stage of the evolutionary algorithm.

[0062] Exemplarily, after the segmented scaling factor is determined using preset coefficients and iteration parameters, before the differential algorithm iterates, the current iteration parameters of the differential algorithm are determined, the stage of the evolutionary algorithm is determined based on the current iteration parameters of the differential algorithm, and the segmented scaling factor corresponding to the evolutionary algorithm at this stage is determined, and the segmented scaling factor is brought into the evolutionary algorithm to obtain the corresponding improved evolutionary algorithm.

[0063] Step 206 , optimizing the initial candidate solution group based on the improved evolutionary algorithm to obtain an optimized candidate solution group, and determining the evolutionary optimal group corresponding to the optimized candidate solution group based on the classifier model.

[0064] Optionally, the initial candidate solution group is optimized by using an improved evolutionary algorithm, which may include perturbing and reorganizing the initial candidate solution group to update the initial candidate solution group and obtain an optimized candidate solution group. And for each group of candidate solutions in the optimized candidate solution group, the fitness function is used to evaluate the effect value corresponding to the group of candidate solutions. Multiple effect values ​​corresponding to the optimized candidate solution group are obtained, and the multiple effect values ​​corresponding to the optimized candidate solution group are compared to determine the highest effect value among the multiple effect values ​​corresponding to the optimized candidate solution group, and the candidate solution corresponding to the highest effect value is used as the evolutionary optimal group.

[0065] Step 208 , determining an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimizing the driving decision model according to the optimized classifier.

[0066] Optionally, the effect value corresponding to the evolutionary optimal group is compared with the effect value corresponding to the initial optimal group to determine the target optimal group. The classifier model is retrained using the key parameters corresponding to the target optimal group to obtain an optimized classifier, and the optimized classifier is applied to the driving decision model to optimize the driving decision model, thereby improving the driving decision process.

[0067] In the above-mentioned driving decision model optimization method, the initial optimal group corresponding to the initial candidate solution group is determined according to the preset classifier model; the segmented scaling factor is determined by using the preset coefficient and the iteration parameter, and the evolutionary algorithm is improved by using the segmented scaling factor to obtain an improved evolutionary algorithm, which can adapt to the different optimization requirements of each stage of the evolutionary algorithm and enhance the generalization ability; the initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain an optimized candidate solution group, and the evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model, and the optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, which can optimize the classifier parameters and improve the classification performance of the classifier, thereby more accurately classifying the data and reducing misjudgment, and optimizing the driving decision model according to the optimized classifier can improve the driving decision accuracy and decision quality, thereby improving driving safety and decision efficiency.

[0068] In the above embodiment, the classifier model is taken as an SVM classifier model as an example. In the SVM classifier model, the appropriate selection of parameters is the key to ensuring the classification effect, especially for nonlinear vector machines, the width parameter of the radial basis kernel function (RBF) used in the SVM classifier model is Controls the radial range of the RBF function, which in turn affects the complexity of the sample data of the SVM classifier model. When the width parameter is larger, the influence range of the RBF kernel function becomes wider, and the influence of each support vector on the SVM classifier model decreases, and the SVM classifier model may become simpler, which may not effectively capture the complexity of the sample data, resulting in underfitting. The smaller the value, the narrower the influence range of the RBF kernel function, and the influence of each support vector on the SVM classifier model increases, so the SVM classifier model may become better adapted to the complex characteristics of the sample data, but it may also lead to overfitting. Therefore, the width parameter is introduced in the candidate solution. .

[0069] In addition, in order to solve the problem that the characteristics of a small part of the sample data domain and the overall sample data may not match in practical problems, a penalty factor C is introduced in the candidate solution. The size of the penalty factor C is directly related to the fault tolerance and overfitting risk of the SVM classifier model. Appropriately increasing the penalty factor C can help reduce misclassification, but excessive increase may cause the model to overfit the training data; while a penalty factor C that is too small will relax the classification rules, increase the misclassification rate, and even cause underfitting problems. Specifically, the penalty factor C is used to control the tolerance of the SVM classifier model to misclassification. When the penalty factor C is smaller, the SVM classifier model allows more misclassifications, the model will be more relaxed, and the decision boundary may be smoother, but it may also cause underfitting. When the penalty factor C is larger, the penalty for misclassification increases, the model will try to correctly classify all sample data, and the decision boundary may become very complex, which may cause the model to overfit.

[0070] In this embodiment, the width parameter is introduced into the candidate solution and penalty operator C, and adopt the segmented scaling factor to improve the evolutionary algorithm to obtain the improved evolutionary algorithm. According to the improved evolutionary algorithm, the initial candidate solution group is optimized, and the width parameter corresponding to the optimal evolutionary group is automatically searched and determined. Combined with the penalty operator C, the width parameter corresponding to the target optimal group is determined and penalty operator C. The improved evolutionary algorithm can automatically adjust the width parameter through continuous iterative evolution. and penalty operator C until the optimal target group is found. The optimal target group is used as the parameter configuration with the best performance of the SVM classifier model to optimize the SVM classifier model, which can avoid the tediousness and subjectivity of manual parameter adjustment and greatly improve the efficiency and accuracy of SVM classifier model parameter optimization. Optimizing the SVM classifier model using the optimal target group can improve the classification performance of the SVM classifier model, and construct and optimize the driving decision model based on the optimized SVM classifier model, so that the driving decision model can accurately identify and predict the behavior patterns in various driving scenarios, provide more reliable driving decisions for the unmanned driving system, and promote the safety and reliability of unmanned driving technology.

[0071] Because traditional evolutionary algorithms overemphasize search capabilities, they may stop searching too early in the process of finding solutions and fail to find better solutions; traditional evolutionary algorithms may also lead to premature convergence of evolutionary algorithms, that is, traditional evolutionary algorithms are too focused on the current solution, resulting in failure to explore more possible solutions, thus missing the optimal solution. Therefore, this application divides the evolutionary algorithm into multiple stages, and determines the corresponding segmented scaling factors for different stages. . Among them, the segment scaling factor The value range is .

[0072] In an exemplary embodiment, the segment scaling factor is determined using a preset coefficient and an iteration parameter. The method comprises: processing the evolutionary algorithm in stages to obtain a multi-stage evolutionary algorithm; determining the preset coefficients corresponding to the evolutionary algorithm in each stage, and determining the segmented scaling factors corresponding to the evolutionary algorithm in each stage according to the preset coefficients and iteration parameters corresponding to the evolutionary algorithm in each stage. .

[0073] Exemplarily, the iteration parameters may include the current iteration number and the maximum iteration number of the evolutionary algorithm. The process of the evolutionary algorithm is divided into multiple stages according to the current iteration number t and the maximum iteration number T of the evolutionary algorithm. For example, the process of the evolutionary algorithm can be divided into the early iteration, the middle iteration and the late iteration.

[0074] Optional, when When , it is determined that the process of the evolutionary algorithm is in the early stage of iteration. When , it is determined that the process of the evolutionary algorithm is in the middle of the iteration. When , it is determined that the process of the evolutionary algorithm is in the late iteration. When the evolutionary algorithm is in the early iteration, in order to enhance the global search ability of the evolutionary algorithm and the diversity of candidate solutions, and further improve the accuracy of the optimal solution, the segmented scaling factor of the early iteration of the evolutionary algorithm is Set to:

[0075]

[0076] Among them, t is the current iteration number of the evolutionary algorithm, T is the maximum iteration number of the evolutionary algorithm, and the preset coefficients in the early stage of the evolutionary algorithm iteration are 0.5 and 4.5.

[0077] When the evolutionary algorithm is in the middle of the iteration, in order to gradually weaken the global search ability of the evolutionary algorithm and gradually transition to local development ability, the segmented scaling factor of the evolutionary algorithm in the middle of the iteration is Set to:

[0078]

[0079] In the late iteration of the evolutionary algorithm, it is necessary to have a strong local development capability to improve the optimization accuracy of the evolutionary algorithm. Therefore, when the evolutionary algorithm is in the late iteration, the segmented scaling factor of the evolutionary algorithm in the late iteration is Set to:

[0080]

[0081] In this embodiment, the evolutionary algorithm is divided into the early stage of iteration, the middle stage of iteration and the late stage of iteration, and the corresponding segmented scaling factors are determined for different stages. , so that the evolutionary algorithm can enhance the diversity of candidate solutions in the early iteration, thereby enhancing the global search ability of the evolutionary algorithm, and in the later iteration of the evolutionary algorithm, it focuses on improving the local development ability of the evolutionary algorithm. Furthermore, the evolutionary algorithm is improved by using a segmented scaling factor to obtain an improved evolutionary algorithm, which can better balance the early search and later development capabilities of the improved evolutionary algorithm, avoid premature stopping or over-concentration of the improved evolutionary algorithm, and thus improve the optimization performance of the improved evolutionary algorithm and find a more accurate optimal solution.

[0082] In an exemplary embodiment, before determining the initial optimal group corresponding to the initial candidate solution group according to a preset classifier model, it also includes: obtaining the number of candidate solution groups; generating dimension parameters for each candidate solution in the candidate solution group according to the number of candidate solution groups and the generation algorithm; generating an initial candidate solution group according to the dimension parameters of each candidate solution; wherein each candidate solution in the initial candidate solution group is a key parameter of the classifier model.

[0083] Exemplarily, the number of candidate solution groups is determined to be ND, and according to the number of candidate solution groups being ND and the generation algorithm shown below, the dimension parameters of each candidate solution in the candidate solution group are generated, and then ND candidate solutions are randomly generated:

[0084]

[0085] in, is the dimension parameter (eigenvalue) of the i-th candidate solution in the candidate solution group in the j-th dimension, for At the lower boundary of the jth dimension, for At the upper boundary of the jth dimension, is a random number between [0,1]. Optional, based on Get the initial candidate solution group .

[0086] In this embodiment, randomly generating candidate solutions can provide diversification of the initial candidate solution group, which helps to prevent the evolutionary algorithm from falling into a local optimal solution, thereby improving the classification performance of the classifier model.

[0087] In an exemplary embodiment, an initial candidate solution group is optimized based on an improved evolutionary algorithm to obtain an optimized candidate solution group, including: performing a perturbation operation on the initial candidate solution group according to a segmented scaling factor to obtain a first improved candidate solution group; and performing a recombination operation on the first improved candidate solution group and the initial candidate solution group to obtain a second improved candidate solution group.

[0088] Exemplarily, three completely different candidate solutions are randomly selected from the initial candidate solution group, and the three completely different candidate solutions randomly selected are perturbed according to the piecewise scaling factor:

[0089]

[0090] in, , and These are three completely different candidate solutions selected from the initial group of candidate solutions. is the piecewise scaling factor, The value of is determined specifically according to the process of the above-mentioned evolutionary algorithm.

[0091] After perturbing the initial candidate solution group according to the piecewise scaling factor, the first improved candidate solution group is obtained. . Improve the first candidate solution group and the initial candidate solution group The reorganization operation includes: comparing the initial candidate solution group and the first improved candidate solution group The advantages and disadvantages of each candidate solution in the solution are evaluated, and the better candidate solutions are retained to obtain the second improved candidate solution group. . For the first improved candidate solution group and the initial candidate solution group Perform a reorganization operation to obtain the second improved candidate solution group The process is as follows:

[0092]

[0093] in, for The random value selected for the jth dimension (dimension parameter), is the crossover probability, The value range is [0,1].

[0094] In this embodiment, the initial candidate solution group Performing perturbations can enable the evolutionary algorithm to more optimally explore the candidate solution space; improving the first candidate solution group and the initial candidate solution group Perform a reorganization operation to combine the first improved candidate solution group and the initial candidate solution group The advantages of the initial candidate solution group are improved, and the accuracy of determining the optimal solution is improved. Updating and adjusting can also enable the evolutionary algorithm to better adapt to different input data, improve the accuracy of the evolutionary algorithm, and thus improve the flexibility and applicability of the classifier.

[0095] In the previous exemplary embodiment, the method further includes: determining a second improved candidate solution group The corresponding effect value and initial candidate solution group Corresponding effect value; According to the second improved candidate solution group The corresponding effect value and initial candidate solution group The corresponding effect value for the second improved candidate solution group and the initial candidate solution group Perform screening operations to obtain the optimized candidate solution group.

[0096] Exemplarily, according to the second improved candidate solution group The corresponding effect value and initial candidate solution group The corresponding effect value for the second improved candidate solution group and the initial candidate solution group The process of screening operation is to obtain the optimized candidate solution group as follows:

[0097]

[0098] in, is the second improved candidate solution group The corresponding effect size, is the initial candidate solution group The corresponding effect size.

[0099] In this embodiment, the second improved candidate solution group The corresponding effect value and initial candidate solution group By comparing the corresponding effect values, the second improved candidate solution group can be objectively judged. and the initial candidate solution group The advantages and disadvantages of each candidate solution in the optimization process are analyzed to determine the optimized candidate solution group. , which can further improve the accuracy of determining the optimal solution.

[0100] In an exemplary embodiment, as shown in FIG3 , determining an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimizing the driving decision model according to the optimized classifier includes steps 302 to 306 . Among them:

[0101] Step 302, obtaining the effect value corresponding to the evolutionary optimal group and the effect value corresponding to the initial optimal group, and comparing the effect value corresponding to the evolutionary optimal group and the effect value corresponding to the initial optimal group to obtain the target optimal group.

[0102] Step 304: Use the target optimal group to adjust the classifier model to obtain an optimized classifier.

[0103] Step 306: Optimize the driving decision model according to the optimized classifier.

[0104] Exemplarily, the initial candidate solution group is optimized based on the improved evolutionary algorithm to obtain the optimized candidate solution group Afterwards, for the optimized candidate solution group For each candidate solution in , the fitness function is used to evaluate the effect value corresponding to the candidate solution. The optimized candidate solution group is obtained The corresponding multiple effect values ​​are grouped into the optimized candidate solutions Compare the corresponding multiple effect values ​​to determine the optimized candidate solution group The highest effect value among the corresponding multiple effect values, and the candidate solution corresponding to the highest effect value is taken as the evolutionary optimal group.

[0105] The effect value corresponding to the evolutionary optimal group is compared with the effect value corresponding to the initial optimal group, and the larger effect value corresponding to the evolutionary optimal group and the initial optimal group is determined, and the optimal group with the larger effect value is used as the target optimal group. The parameters in the target optimal group are used to adjust the classifier model. For example, the width parameter in the target optimal group is The penalty operator C is loaded into the classifier and retrained to obtain an optimized classifier to ensure that the classifier model can make full use of the optimized parameters and improve the adaptability of the optimized classifier in different driving scenarios. After obtaining the optimized classifier, the driving decision model is re-evaluated and optimized through the optimized classifier.

[0106] In this embodiment, the optimized classifier can more accurately identify and classify various situations in the driving environment through improved parameters, thereby improving the classification performance and accuracy of the classifier. Re-evaluating and optimizing the driving decision model using the optimized classifier can improve the accuracy of driving decisions, reduce potential safety risks, and improve driving efficiency.

[0107] In another embodiment, a driving decision model optimization method is provided, the method comprising:

[0108] Obtain the number of candidate solution groups; generate dimension parameters for each candidate solution in the candidate solution group according to the number of candidate solution groups and the generation algorithm; generate an initial candidate solution group according to the dimension parameters of each candidate solution; wherein each individual in the initial candidate solution group is a key parameter of the classifier model.

[0109] According to the preset classifier model, the initial optimal group corresponding to the initial candidate solution group is determined.

[0110] The evolutionary algorithm is processed in stages to obtain a multi-stage evolutionary algorithm; the preset coefficients corresponding to the evolutionary algorithm in each stage are determined, and the segmented scaling factors corresponding to the evolutionary algorithm in each stage are determined according to the preset coefficients and iteration parameters corresponding to the evolutionary algorithm in each stage; the evolutionary algorithm is improved by using the segmented scaling factors to obtain an improved evolutionary algorithm.

[0111] The initial candidate solution group is perturbed according to the piecewise scaling factor to obtain a first improved candidate solution group; the first improved candidate solution group and the initial candidate solution group are reorganized to obtain a second improved candidate solution group.

[0112] Determine the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group; according to the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group, perform a screening operation on the second improved candidate solution group and the initial candidate solution group to obtain an optimized candidate solution group.

[0113] The evolutionary optimal group corresponding to the optimized candidate solution group is determined based on the classifier model.

[0114] Obtain the effect value corresponding to the evolutionary optimal group and the effect value corresponding to the initial optimal group, and compare the effect value corresponding to the evolutionary optimal group with the effect value corresponding to the initial optimal group to obtain the target optimal group; use the target optimal group to adjust the classifier model to obtain an optimized classifier; optimize the driving decision model according to the optimized classifier.

[0115] In this embodiment, randomly generating candidate solutions can provide diversification of the initial candidate solution group, which helps to avoid the evolutionary algorithm from falling into the local optimal solution, thereby improving the classification performance of the classifier model. The evolutionary algorithm is improved by adopting a segmented scaling factor to obtain an improved evolutionary algorithm, which can better balance the early search and later development capabilities of the improved evolutionary algorithm, and can avoid the improved evolutionary algorithm from stopping too early or over-concentrating, thereby improving the optimization performance of the improved evolutionary algorithm and finding a more accurate optimal solution. The initial candidate solution group is updated and adjusted by the improved evolutionary algorithm to obtain an optimized candidate solution group, which can improve the accuracy of the evolutionary algorithm and obtain a more accurate target optimal group. The optimized classifier is obtained according to the target optimal group, so that the optimized classifier can more accurately identify and classify various situations in the driving environment, and improve the classification performance and accuracy of the classifier. The driving decision model is re-evaluated and optimized by the optimized classifier, which can improve the accuracy of driving decisions, reduce potential safety risks, and improve driving efficiency.

[0116] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0117] Based on the same inventive concept, the embodiment of the present application also provides a driving decision model optimization device for implementing the driving decision model optimization method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more driving decision model optimization device embodiments provided below can refer to the limitations of the driving decision model optimization method above, and will not be repeated here.

[0118] In an exemplary embodiment, as shown in FIG4 , a driving decision model optimization device is provided, comprising: a first determination module 402 , an improvement module 404 , a second determination module 406 and an optimization module 408 , wherein:

[0119] A first determination module 402 is used to determine an initial optimal group corresponding to an initial candidate solution group according to a preset classifier model;

[0120] An improvement module 404 is used to determine a segmented scaling factor using a preset coefficient and an iteration parameter, and improve the evolutionary algorithm using the segmented scaling factor to obtain an improved evolutionary algorithm;

[0121] The second determination module 406 is used to optimize the initial candidate solution group based on the improved evolutionary algorithm to obtain an optimized candidate solution group, and determine the evolutionary optimal group corresponding to the optimized candidate solution group based on the classifier model;

[0122] The optimization module 408 is used to determine an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimize the driving decision model according to the optimized classifier.

[0123] In an exemplary embodiment, the improvement module 404 is also used to process the evolutionary algorithm in stages to obtain a multi-stage evolutionary algorithm; determine the preset coefficients corresponding to the evolutionary algorithm at each stage, and determine the segmented scaling factors corresponding to the evolutionary algorithm at each stage based on the preset coefficients and iteration parameters corresponding to the evolutionary algorithm at each stage.

[0124] In an exemplary embodiment, the optimization module 408 is also used to obtain the effect value corresponding to the evolutionary optimal group and the effect value corresponding to the initial optimal group, and compare the effect value corresponding to the evolutionary optimal group with the effect value corresponding to the initial optimal group to obtain the target optimal group; use the target optimal group to adjust the classifier model to obtain an optimized classifier; and optimize the driving decision model according to the optimized classifier.

[0125] In an exemplary embodiment, the second determination module 406 is further configured to perform a perturbation operation on the initial candidate solution group according to the segmented scaling factor to obtain a first improved candidate solution group; and perform a reorganization operation on the first improved candidate solution group and the initial candidate solution group to obtain a second improved candidate solution group.

[0126] In an exemplary embodiment, the second determination module 406 is also used to determine the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group; according to the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group, the second improved candidate solution group and the initial candidate solution group are screened to obtain an optimized candidate solution group.

[0127] In an exemplary embodiment, the driving decision model optimization device further includes:

[0128] The initial determination module is used to obtain the number of candidate solution groups; generate dimension parameters for each candidate solution in the candidate solution group according to the number of candidate solution groups and the generation algorithm; generate an initial candidate solution group according to the dimension parameters of each candidate solution; wherein each individual in the initial candidate solution group is a key parameter of the classifier model.

[0129] Each module in the above driving decision model optimization device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0130] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG5. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device can be used to store segmented scaling factor data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a driving decision model optimization method is implemented.

[0131] Those skilled in the art will understand that the structure shown in FIG. 5 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0132] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0133] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0134] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0135] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0136] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0137] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0138] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A driving decision model optimization method, characterized in that: The method comprises: According to the preset classifier model, determine the initial optimal group corresponding to the initial candidate solution group; Using a preset coefficient and an iteration parameter to determine a segmented scaling factor, and using the segmented scaling factor to improve the evolutionary algorithm to obtain an improved evolutionary algorithm; Optimizing the initial candidate solution group based on the improved evolutionary algorithm to obtain an optimized candidate solution group, and determining the evolutionary optimal group corresponding to the optimized candidate solution group based on the classifier model; An optimized classifier is determined according to the evolutionary optimal group and the initial optimal group, and a driving decision model is optimized according to the optimized classifier.

2. The method according to claim 1, characterized in that The step of using the preset coefficient and the iterative parameter to determine the segmented scaling factor comprises: Processing the evolutionary algorithm in stages to obtain a multi-stage evolutionary algorithm; Determine the preset coefficients corresponding to the evolutionary algorithm at each stage, and determine the segmented scaling factors corresponding to the evolutionary algorithm at each stage according to the preset coefficients and iteration parameters corresponding to the evolutionary algorithm at each stage.

3. The method according to claim 1, characterized in that Determining an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimizing the driving decision model according to the optimized classifier comprises: Obtaining the effect value corresponding to the evolutionary optimal group and the effect value corresponding to the initial optimal group, and comparing the effect value corresponding to the evolutionary optimal group with the effect value corresponding to the initial optimal group to obtain a target optimal group; The classifier model is adjusted using the target optimal group to obtain an optimized classifier; The driving decision model is optimized according to the optimized classifier.

4. The method according to claim 1, characterized in that: The optimizing the initial candidate solution group based on the improved evolutionary algorithm to obtain the optimized candidate solution group includes: Performing a perturbation operation on the initial candidate solution group according to the piecewise scaling factor to obtain a first improved candidate solution group; The first improved candidate solution group and the initial candidate solution group are reorganized to obtain a second improved candidate solution group.

5. The method according to claim 4, characterized in that The method further comprises: Determine the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group; According to the effect value corresponding to the second improved candidate solution group and the effect value corresponding to the initial candidate solution group, a screening operation is performed on the second improved candidate solution group and the initial candidate solution group to obtain an optimized candidate solution group.

6. The method according to claim 1, characterized in that Before determining the initial optimal group corresponding to the initial candidate solution group according to the preset classifier model, the method further includes: Obtaining the number of candidate solution groups; generating dimension parameters for each candidate solution in the candidate solution group according to the number of candidate solution groups and a generation algorithm; The initial candidate solution group is generated according to the dimension parameters of each candidate solution; wherein each individual in the initial candidate solution group is a key parameter of the classifier model.

7. A driving decision model optimization device, characterized in that: The device comprises: A first determination module is used to determine an initial optimal group corresponding to an initial candidate solution group according to a preset classifier model; An improvement module, used to determine a segmented scaling factor using a preset coefficient and an iteration parameter, and to improve the evolutionary algorithm using the segmented scaling factor to obtain an improved evolutionary algorithm; A second determination module is used to optimize the initial candidate solution group based on the improved evolutionary algorithm to obtain an optimized candidate solution group, and determine the evolutionary optimal group corresponding to the optimized candidate solution group based on the classifier model; The optimization module is used to determine an optimized classifier according to the evolutionary optimal group and the initial optimal group, and optimize the driving decision model according to the optimized classifier.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images