A Simulation Parameter Decision-Making Method, Device, and Medium Based on Deep Learning Representations

Abstract

The present application discloses a simulation parameter decision-making method, device and medium based on deep learning representation, which relates to the field of simulation parameter decision-making. The method includes: obtaining simulation data of historical vehicle design modeling, and extracting vehicle design parameters and vehicle performance parameters from the simulation data; applying the CFS method to screen out key vehicle design parameters from the design parameter set; using the key vehicle design parameters as network inputs and the vehicle performance parameters as network outputs to train the neural network KAN, and then obtaining a non-linear equation set between the key vehicle design parameters and each vehicle performance parameter; taking that each vehicle performance parameter meets the set target as the parameter optimization target, and using a multi-objective optimization algorithm to solve the non-linear equation set to obtain the optimal vehicle design parameter values. Applying the method of the present invention can solve the problems of inaccurate decision results and low efficiency caused by a long time period existing in the existing simulation parameter decision-making methods.

Description

Technical Field

[0001] This application relates to the field of simulation parameter decision-making, and particularly to a simulation parameter decision-making method, device, and medium based on deep learning representation. Background Technique

[0002] In recent years, the automotive industry has actively adapted to the industrial development trend, strengthened the innovation main body status of automotive enterprises, and become a highland leading the global automotive industry's innovative development. Among them, the new energy vehicle industry has achieved positive results, injecting strong impetus into the electrification transformation of the global automotive industry. It integrates multiple transformative technologies and has become an important carrier for the implementation of new technologies.

[0003] The research and development of new energy vehicles is an iterative process of design, testing, and optimization. The interaction of complex multi-physical systems in vehicles poses challenges to the development of new energy vehicles. Modeling and simulation technologies provide convenient means for checking and optimizing the complex interactions within electric vehicles. The vehicle development at the system level of simulation software involves multiple aspects such as vehicle performance, performance integration, energy management, PE performance development, thermal management, technology planning, new energy system development, powertrain, and fuel cell simulation. Reducing the number of required experiments through simulation technology has become an auxiliary development method for vehicle enterprises to reduce development costs, shorten product launch times, and improve product quality.

[0004] Traditional simulation parameter decision-making methods can find the optimal combination of design parameters by performing a systematic search within a predefined range of design parameters. However, this method has a large computational overhead. There are also control strategies that enable the control system to quickly respond to changes in the external environment through real-time feedback and parameter adjustment, that is, the error is eliminated based on the error. For example, PID Tuner adjusts the output of the system through the combination of three links: proportional (P), integral (I), and derivative (D) to achieve tracking of the reference input and suppression of disturbances. In addition, a more common simulation parameter decision-making method is to list all parameters flatly. Engineers rely on their own design experience or existing established rules to define design parameters and gradually adjust them according to the simulation results until a reasonable result is obtained. However, the existing simulation decision-making methods have some disadvantages: (1) For the method of real-time feedback and parameter adjustment, on the one hand, it is restricted by the particularity of the simulation system. For example, for a non-linear system, this method may be difficult to provide the best control effect. On the other hand, it is sensitive to both noise and external disturbances, especially the derivative term, which may cause the output of the simulation system to become unstable when facing high-frequency noise. In addition, there is also the problem of integral saturation (integral overflow). When the simulation system cannot reach the set point for a long time, the integral term may accumulate too much, resulting in controller saturation and affecting the control effect. (2) For simulation systems that cannot use the real-time feedback adjustment method, if relying on the most traditional manual parameter adjustment method, if engineers lack prior experience, it takes a long time to repeatedly adjust and optimize multiple vehicle design parameters through custom settings of vehicle design parameters to achieve the best performance.

[0005] Therefore, the existing simulation parameter decision-making methods have problems of inaccurate decision results and low efficiency due to the long time cycle. Summary of the Invention

[0006] The purpose of this application is to provide a simulation parameter decision-making method, device, and medium based on deep learning representation, which can solve the problems of inaccurate decision results and low efficiency due to the long time cycle existing in the existing simulation parameter decision-making methods.

[0007] To achieve the above purpose, this application provides the following solutions.

[0008] In the first aspect, this application provides a simulation parameter decision-making method based on deep learning representation, including the following steps:

[0009] Obtain the simulation data of historical vehicle design modeling, extract vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set;

[0010] Apply the CFS method to screen out the key vehicle design parameters from the design parameter set to obtain a key design parameter set;

[0011] Using each key vehicle design parameter in the set of key design parameters as the network input, and using the corresponding vehicle performance parameters in the set of performance parameters as the network output, train the neural network KAN to obtain the trained neural network KAN;

[0012] According to the network structure of the trained neural network KAN, obtain a system of non-linear equations between the key vehicle design parameters and the vehicle performance parameters; one non-linear equation in the system of non-linear equations is a non-linear relationship formula between a vehicle performance parameter and the key vehicle design parameters;

[0013] Taking the satisfaction of each vehicle performance parameter with the set target as the parameter optimization goal, use a multi-objective optimization algorithm to solve the system of non-linear equations to obtain the optimal vehicle design parameter values.

[0014] In a second aspect, the present application provides a simulation parameter decision-making device based on deep learning representation, including the following modules:

[0015] A simulation data acquisition module, configured to acquire simulation data of historical vehicle design modeling, extract vehicle design parameters and vehicle performance parameters from the simulation data to obtain a set of design parameters and a set of performance parameters;

[0016] A key vehicle design parameter screening module, configured to screen out key vehicle design parameters from the set of design parameters by applying the CFS method to obtain a set of key design parameters;

[0017] A neural network KAN training module, configured to use each key vehicle design parameter in the set of key design parameters as the network input, and use the corresponding vehicle performance parameters in the set of performance parameters as the network output, train the neural network KAN to obtain the trained neural network KAN;

[0018] A non-linear equation system construction module, configured to obtain a system of non-linear equations between the key vehicle design parameters and the vehicle performance parameters according to the network structure of the trained neural network KAN; one non-linear equation in the system of non-linear equations is a non-linear relationship formula between a vehicle performance parameter and the key vehicle design parameters;

[0019] A solving module, configured to take the satisfaction of each vehicle performance parameter with the set target as the parameter optimization goal, use a multi-objective optimization algorithm to solve the system of non-linear equations to obtain the optimal vehicle design parameter values.

[0020] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned simulation parameter decision-making method based on deep learning representation.

[0021] Fourthly, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-described simulation parameter decision method based on deep learning representation is implemented.

[0022] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0023] The present application provides a simulation parameter decision method, device and medium based on deep learning representation, which obtains simulation data of historical vehicle design modeling, extracts vehicle design parameters input to the simulation system and vehicle performance parameters output by the simulation system from the simulation data; applies the CFS method to screen out key vehicle design parameters from the extracted vehicle design parameters; uses the key vehicle design parameters as the network input and the output vehicle performance parameters as the network output to train the neural network KAN; obtains a non-linear equation set between the key vehicle design parameters and the vehicle performance parameters according to the network structure of the trained neural network KAN; takes the satisfaction of each vehicle performance parameter with the set target as the parameter optimization target, and uses a multi-objective optimization algorithm to solve the non-linear equation set to obtain the optimal vehicle design parameter value. The present invention takes into account a large amount of expert experience and design knowledge contained in the existing vehicle established parameter settings and data generated by multiple simulations, which has certain significance for vehicle development. Therefore, the present invention makes full use of these existing vehicle simulation data, excavates and reuses the knowledge and rules implicit in the relationship between vehicle design parameters and vehicle performance, so as to reduce the actual number of simulations, thereby improving the iterative design efficiency and reducing the simulation development cost. In addition, key vehicle design parameters are extracted from numerous vehicle design parameters, reducing redundant design parameters and improving the computational efficiency of simulation parameter decision-making. In addition, the present invention uses key vehicle design parameters and vehicle performance parameters to train the neural network KAN, represents the simulation process with a deep learning algorithm, transforms the simulation parameter recommendation problem into a multi-objective optimization problem, and solves the global optimal solution of the simulation process. Since the training of the neural network KAN is not restricted by the particularity of the simulation system, and the spline function activation mechanism in the neural network enables it to more accurately capture the non-linear relationship between vehicle design parameters and vehicle performance parameters, improving the fitting accuracy. Therefore, compared with the existing simulation decision methods, the present invention not only ensures the efficiency of simulation parameter decision-making but also ensures the accuracy of simulation parameter decision-making. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0025] Figure 1 It is an application environment diagram of a simulation parameter decision-making method based on deep learning representation in an embodiment of the present application;

[0026] Figure 2 It is a schematic flowchart of a simulation parameter decision-making method based on deep learning representation provided in an embodiment of the present application;

[0027] Figure 3 It is a schematic diagram of the network structure of the neural network KAN provided in an embodiment of the present application;

[0028] Figure 4 It is a schematic diagram of the functional modules of a simulation parameter decision-making device based on deep learning representation provided in an embodiment of the present application;

[0029] Figure 5 It is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Detailed implementation manners

[0030] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0031] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the drawings and specific implementation manners.

[0032] The simulation parameter decision-making method based on deep learning representation provided in the embodiments of the present application can be applied to, for example Figure 1In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be set separately, integrated on the server, or placed on the cloud or other servers. The terminal can send the simulation data of the historical vehicle design modeling to the server. After the server receives the simulation data of the historical vehicle design modeling, the server extracts the vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set; uses the CFS method to screen out the key vehicle design parameters from the design parameter set to obtain a key design parameter set; uses each key vehicle design parameter in the key design parameter set as the network input and each vehicle performance parameter in the performance parameter set as the network output to train the neural network KAN to obtain the trained neural network KAN; obtains the non-linear equations between the key vehicle design parameters and each vehicle performance parameter according to the network structure of the trained neural network KAN; uses the satisfaction of each vehicle performance parameter with the set target as the parameter optimization target, and uses the multi-objective optimization algorithm to solve the non-linear equations to obtain the optimal vehicle design parameter value. The server can feedback the obtained optimal vehicle design parameter value to the terminal. In addition, in some embodiments, the simulation parameter decision method based on deep learning representation can also be implemented separately by the server or the terminal. For example, the terminal can directly calculate the optimal vehicle design parameter value for the simulation data of the historical vehicle design modeling, or the server can obtain the simulation data of the historical vehicle design modeling from the data storage system and calculate the optimal vehicle design parameter value.

[0033] Among them, the terminal can be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0034] In an exemplary embodiment, as Figure 2 shown, a simulation parameter decision method based on deep learning representation is provided. This method is executed by a computer device, and can be specifically executed separately by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, taking this method applied to Figure 1 the server in as an example for illustration, it includes the following steps 101 to step 105.

[0035] Step 101, obtain the simulation data of the historical vehicle design modeling, and extract the vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set.

[0036] Vehicle design parameters include dimensional parameters (such as vehicle length, vehicle width, vehicle height, wheelbase, track width, minimum ground clearance, etc.), weight parameters (such as curb weight, gross weight, payload, etc.), body parameters (such as body type, number of doors, number of seats, etc.), powertrain parameters (such as engine type, engine displacement, engine layout, drive type, etc.), suspension system parameters (such as front suspension type, rear suspension type, etc.), braking system parameters (such as front brake type, rear brake type, etc.), and so on.

[0037] Vehicle performance parameters include power performance parameters (such as the maximum power of the engine, the maximum torque of the engine, the time required for the vehicle to accelerate from standstill to 100 km / h, the maximum speed, etc.), braking performance parameters (such as the 100 - 0 km / h braking distance, braking system response time, etc.), fuel performance parameters (such as combined fuel consumption, driving range, etc.), and so on.

[0038] Step 102, apply the CFS method to screen out the key vehicle design parameters from the design parameter set to obtain the key design parameter set. The key vehicle design parameters are part of the vehicle design parameters screened out from the design parameter set using the CFS method.

[0039] Step 103, use each key vehicle design parameter in the key design parameter set as the network input, and each vehicle performance parameter in the performance parameter set as the network output to train the neural network KAN to obtain the trained neural network KAN.

[0040] By introducing the neural network KAN improved by MLP to represent the simulation process, taking the extracted vehicle performance parameters as the output of the neural network KAN and the key vehicle design parameters of the simulation input as the input of the neural network KAN, train the neural network KAN.

[0041] The neural network KAN has learnable activation functions on the edges between nodes. These activation functions are constructed using basis functions, so the activation functions do not need to be specified in the training structure. By introducing the Kolmogorov - Arnold representation theorem (KAT) and B - spline functions, the neural network KAN improves its data - processing method in each training iteration, thus continuously improving its accuracy and efficiency. Figure 3 For the network structure of a neural network KAN, x is the m inputs of the neural network KAN, and KAN(x) is the n outputs of the neural network KAN. 、 are non - linear learnable activation functions. Use ADAM as the optimizer and the root - mean - square error RMSE as the loss function. The loss function is as follows.

[0042] 。

[0043] Among them, is the calculated output value of the neural network KAN, that is, the vehicle performance parameters output by the network; , ,..., ,..., is the true simulation output result in the simulation data, that is, the vehicle performance parameters in the simulation data; i refers to the i-th vehicle performance parameter; n is the number of observed vehicle performance parameters.

[0044] After the neural network KAN is trained, the edges between some neurons are lower than the threshold and will be pruned, and the communication of the neural network KAN will be more efficient. Based on the particularity of the neural network KAN, the structure of the network after training can be extracted into a symbolic formula and used for the optimization of the subsequent simulation system design parameters.

[0045] Step 104, obtain a non-linear equation set between the key vehicle design parameters and each vehicle performance parameter according to the network structure of the trained neural network KAN; one non-linear equation in the non-linear equation set is a non-linear relationship formula between a vehicle performance parameter and each key vehicle design parameter.

[0046] Use the symbolic formula obtained by the trained neural network KAN to recommend the simulation design parameters of the simulation system, which is due to the KAT theorem on which the neural network KAN is based. Therefore, for each vehicle performance parameter, the trained neural network KAN can extract a non-linear formula calculated by the key vehicle design parameters, and thus, a set of non-linear equation sets are obtained for each vehicle performance parameter.

[0047] Step 105, taking that each vehicle performance parameter meets the set target as the parameter optimization goal, use the multi-objective optimization algorithm to solve the non-linear equation set to obtain the optimal design parameter values, that is, the optimal numerical combination of the key vehicle design parameters.

[0048] By implementing the above steps 101 to 105, considering that a large amount of expert experience and design knowledge are contained in the existing vehicle established parameter settings and data generated by multiple simulations, which is of certain significance to vehicle development, this application makes full use of these existing vehicle simulation data, mines and reuses the knowledge and rules implicit in the relationship between vehicle design parameters and vehicle performance, so as to reduce the actual number of simulations, improve the iterative design efficiency, and reduce the simulation development cost. In addition, key vehicle design parameters are extracted from numerous vehicle design parameters, reducing redundant design parameters and improving the computational efficiency of simulation parameter decision-making. In addition, this application uses the key vehicle design parameters and the vehicle performance parameters output by the simulation to train the neural network KAN, characterizes the simulation process with a deep learning algorithm, transforms the simulation parameter recommendation problem into a multi-objective optimization problem, and solves the global optimal solution of the simulation process. Since the training of the neural network KAN is not restricted by the particularity of the simulation system, and the spline function activation mechanism in the neural network enables it to more accurately capture the non-linear relationship between vehicle design parameters and vehicle performance parameters, improving the fitting accuracy. Therefore, compared with the existing simulation decision-making methods, the present invention ensures both the efficiency and the accuracy of simulation parameter decision-making.

[0049] In another exemplary embodiment of this application, for step 101, vehicle design parameters for simulation input and vehicle performance parameters for simulation output are extracted from historical vehicle design modeling simulation data, and the parameter data is cleaned to eliminate samples with missing values and outliers, getting rid of the influence of abnormal data on the subsequent training process. Then the vehicle performance parameters are standardized and transformed into a unified unit format for subsequent data mining. Therefore, step 101, obtaining the simulation data of historical vehicle design modeling, extracting vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set, specifically includes the following steps.

[0050] (1) Obtaining the simulation data of historical vehicle design modeling.

[0051] (2) Preprocessing the simulation data to obtain preprocessed simulation data.

[0052] (3) Performing standardization processing on the preprocessed simulation data to obtain standardized simulation data.

[0053] (4) Extracting vehicle design parameters and vehicle performance parameters from the standardized simulation data to obtain a design parameter set and a performance parameter set.

[0054] In another exemplary embodiment of the present application, for step 102, the key parameter selection attempts to find the key vehicle design parameters that affect the output of the simulation results and eliminate the redundant vehicle design parameters that have no effect on the simulation results. The vehicle design parameters are mainly screened by simultaneously considering the correlations between vehicle design parameters and between vehicle design parameters and target variables (vehicle performance parameters). The CFS (Correlation-based Feature Selection) method is adopted, which is a heuristic feature selection method based on statistical metrics. This method is based on the following assumption: A good feature subset contains features that are highly correlated with the result but not correlated with each other. The CFS method first calculates the correlation matrices between vehicle design parameters and vehicle performance parameters and between vehicle design parameters from the design parameter set; during the feature selection process, the CFS method aims to find a subset S of vehicle design parameters, where k vehicle design parameters are included. These vehicle design parameters should have a high correlation with the target variable and a low correlation with each other. The CFS method uses an evaluation function (Merit) to evaluate the quality of the subset S of vehicle design parameters, so the Merit (evaluation value) of the subset S of vehicle design parameters composed of k vehicle design parameters can be calculated. The CFS method will traverse all possible subsets of vehicle design parameters and calculate their Merit evaluation values. Finally, the subset of vehicle design parameters with the highest Merit evaluation value is selected as the final set of key design parameters.

[0055] The form of the Merit function is shown as follows.

[0056] ;

[0057] Wherein, is the average value of the correlations between all vehicle design parameters and vehicle performance parameters; is the average value of the correlations between all vehicle design parameters and vehicle design parameters; represents the subset S composed of k vehicle design parameters.

[0058] The CFS method obtains a relatively optimal subset of vehicle design parameters by continuously iteratively updating the subset of vehicle design parameters and calculating their Merit evaluation values. The iterative update of the subset of vehicle design parameters uses best first search to search the space of the subset of vehicle design parameters. Initially, the subset of vehicle design parameters consists of all possible single vehicle design parameters; calculate the estimated values of the vehicle design parameters (represented by the Merit evaluation values), and select the vehicle design parameter with the largest Merit evaluation value to enter subset S. Then select the vehicle design parameter with the second largest Merit evaluation value to enter subset S. If the Merit evaluation value of the subset composed of these two vehicle design parameters is less than the Merit evaluation value of the original subset, then the vehicle design parameter with the second largest Merit evaluation value is not added to subset S. After that, select the next vehicle design parameter, and so on, to obtain the combination of vehicle design parameters that maximizes the Merit evaluation value, which is the key design parameter set. Therefore, in step 102, the CFS method is used to screen out the key vehicle design parameters from the design parameter set to obtain the key design parameter set, which specifically includes:

[0059] (1) Calculate the first correlation between the key vehicle design parameters and the vehicle performance parameters, and the second correlation between the key vehicle design parameters.

[0060] (2) Calculate the Merit evaluation value of each vehicle design parameter in the design parameter set according to the first correlation and the second correlation.

[0061] (3) Screen out the vehicle design parameter with the largest Merit evaluation value as the key vehicle design parameter. Add the screened key vehicle design parameter to the key design parameter set, and calculate the Merit evaluation value of the current key design parameter set; initially, the key design parameter set is an empty set.

[0062] (4) Use the vehicle design parameter with the largest Merit evaluation value in the current remaining design parameter set and all the key vehicle design parameters in the current key design parameter set as a temporary subset, and calculate the Merit evaluation value of the temporary subset; the remaining design parameter set refers to the set composed of the parameters in the design parameter set except the current key vehicle design parameters.

[0063] (5) Judge whether the Merit evaluation value of the temporary subset is greater than the Merit evaluation value of the current key design parameter set.

[0064] (6) If so, use the temporary subset as the current critical design parameter set, and return to the step "Use the vehicle design parameter with the maximum Merit value in the current remaining design parameter set and all critical vehicle design parameters in the current critical design parameter set as the temporary subset".

[0065] (7) If not, use the parameters in the current remaining design parameter set except for the vehicle design parameters added to the temporary subset to form the current remaining design parameter set, and return to the step "Use the vehicle design parameter with the maximum Merit value in the current remaining design parameter set and all critical vehicle design parameters in the current critical design parameter set as the temporary subset".

[0066] (8) Keep doing this until the maximum value of the Merit value of the critical design parameter set is found, and then obtain the final critical design parameter set.

[0067] In another exemplary embodiment of the present application, to prevent the CFS method from falling into a local optimal solution, K-fold cross-validation is performed on the original design parameter set, that is, the original design parameter set is divided into K subsets of equal size. Each time, one of the subsets is selected as the validation set, and the remaining K-1 subsets are used as the training set. This process is repeated K times, and each time a different subset is selected as the validation set. In each iteration, the MLP model is trained using the training set, and the validation set is used to evaluate the performance of the MLP model. The performance evaluation metric is the mean absolute error MAE. According to the results of the cross-validation, analyze the influence of different training sets on the performance of the MLP model. Usually, the training set with the best performance on the validation set is selected and compared with the performance of the critical design parameter set to evaluate the performance of the critical design parameter set. If the performance of the training set with the best performance on the validation set is better than the performance of the critical design parameter set, then the training set with the best performance on the validation set is used as the critical design parameter set.

[0068] Therefore, after the step 102 "Apply the CFS method to screen out the critical vehicle design parameters from the design parameter set to obtain the critical design parameter set" is executed in the present application, the simulation parameter decision method based on deep learning representation further includes: evaluating the critical design parameter set, specifically including the following steps.

[0069] (1) Divide the design parameter set into K subsets.

[0070] (2) Use each subset as the validation set respectively, and the remaining K-1 subsets as the training set.

[0071] (3) Use the training set to train the MLP model to obtain the first trained MLP model.

[0072] (4) Use the validation set to verify the first trained MLP model to obtain the first verification evaluation metric.

[0073] (5) Determine the optimal first verification evaluation index.

[0074] (6) Use the set of key design parameters as the training set to train the MLP model, and obtain the second trained MLP model.

[0075] (7) Use the remaining set of design parameters as the verification set to verify the second trained MLP model, and obtain the second verification evaluation index.

[0076] (8) Determine whether the second verification evaluation index is better than the optimal first verification evaluation index.

[0077] (9) If so, retain the set of key design parameters.

[0078] (10) If not, use the training set corresponding to the optimal first verification evaluation index as the set of key design parameters.

[0079] In another exemplary embodiment of the present application, the problem of solving the optimal vehicle design parameters of the simulation system becomes the problem of seeking the optimal solution of the non-linear equations, that is, it can be solved using multi-objective optimization algorithms such as genetic algorithms and particle swarm algorithms. Therefore, in step 105, taking that each vehicle performance parameter meets the set target as the parameter optimization target, use the multi-objective optimization algorithm to solve the non-linear equations, and obtain the optimal vehicle design parameter values, which specifically include the following steps.

[0080] (1) Taking that each vehicle performance parameter meets the set target as the parameter optimization target, determine the objective function.

[0081] (2) Use the genetic algorithm or particle swarm algorithm to solve the non-linear equations, and obtain the optimal vehicle design parameter values; in the genetic algorithm or particle swarm algorithm, an individual in the population refers to a numerical combination of each key vehicle design parameter that satisfies the non-linear equations and the objective function.

[0082] In this embodiment, first, key vehicle design parameters are selected through a vehicle design parameter mining method, and the performance of the key design parameter set is evaluated through cross-validation. Then, based on these parameters, a deep learning algorithm is used to characterize the simulation process. After transforming the simulation parameter recommendation problem into a multi-objective optimization problem, the global optimal solution of the simulation process is solved to discover the underlying knowledge in the massive simulation data and provide support for the vehicle design parameter decision-making of engineers. Among them, the key vehicle design parameter selection step can reduce the problem of too long model training time caused by redundant parameters, eliminate the design parameters that do not affect the results, and improve the model training efficiency. Cross-validation evaluation can prevent the selected key vehicle design parameters from falling into a local optimum and ensure that the parameters input into the neural network are representative. Using deep learning to solve and characterize the simulation process, a deep learning neural network is trained using key vehicle design parameters, and symbolic formulas are extracted using an interpretable deep learning model, laying a foundation for the solution of the optimal vehicle design parameters. Finally, the symbolic formula obtained by deep learning is used for vehicle design parameter recommendation.

[0083] In this embodiment, deep learning is an end-to-end learning process. The training of the model is not restricted by the particularity of the simulation system, and the spline function activation mechanism in the neural network enables it to more accurately capture the non-linear relationships in the data and improve the fitting accuracy. For the noise and external interference in the data, this application adopts different technical means to improve the noise robustness of the method, such as the data preprocessing method in the early stage. And when dealing with high-dimensional data, the neural network KAN can decompose complex functions, remove noise while retaining key information, and improve the efficiency of subsequent analysis.

[0084] This application also provides an application scenario that applies the above-mentioned simulation parameter decision-making method based on deep learning characterization. Specifically: The simulation parameter decision-making method based on deep learning characterization provided in this embodiment can be applied in the vehicle design scenario. This scenario includes a data collection link, a data processing link, and a vehicle design link; the data collection link is used to obtain the simulation data of historical vehicle design modeling; the data processing link is used to process the simulation data to obtain the optimal vehicle design parameter values; the vehicle design link is used to carry out vehicle design with reference to the optimal vehicle design parameter values. The simulation parameter decision-making method based on deep learning characterization provided in this embodiment belongs to the data processing link.

[0085] Based on the same inventive concept, an embodiment of the present application further provides a simulation parameter decision-making device based on deep learning representation for implementing the simulation parameter decision-making method based on deep learning representation involved above. The implementation solution provided by this device for solving problems is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the simulation parameter decision-making device based on deep learning representation provided below can refer to the limitations on the simulation parameter decision-making method based on deep learning representation in the above text, and will not be repeated here.

[0086] In an exemplary embodiment, as Figure 4 shown, a simulation parameter decision-making device based on deep learning representation is provided, which includes the following modules.

[0087] A simulation data acquisition module M1, configured to acquire simulation data of historical vehicle design modeling, extract vehicle design parameters and vehicle performance parameters from the simulation data, and obtain a design parameter set and a performance parameter set.

[0088] A key vehicle design parameter screening module M2, configured to screen out key vehicle design parameters from the design parameter set by applying the CFS method to obtain a key design parameter set.

[0089] A neural network KAN training module M3, configured to use each key vehicle design parameter in the key design parameter set as a network input, and use the corresponding vehicle performance parameter in the performance parameter set as a network output to train the neural network KAN to obtain a trained neural network KAN.

[0090] A non-linear equation system construction module M4, configured to obtain a non-linear equation system between key vehicle design parameters and each vehicle performance parameter according to the network structure of the trained neural network KAN; one non-linear equation in the non-linear equation system is a non-linear relationship formula between a vehicle performance parameter and each key vehicle design parameter.

[0091] A solving module M5, configured to use the satisfaction of each vehicle performance parameter with a set target as a parameter optimization target, and use a multi-objective optimization algorithm to solve the non-linear equation system to obtain an optimal vehicle design parameter value.

[0092] In an exemplary embodiment, a computer device (i.e., the Figure 5 computer device in) is provided. This computer device can be a server or a terminal, and its internal structure diagram can be as Figure 5As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated 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 is used to store optimal design parameter value data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a simulation parameter decision method based on deep learning representation.

[0093] Those skilled in the art can understand that Figure 5 The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, it implements the above-mentioned simulation parameter decision method based on deep learning representation.

[0094] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by the processor, it implements the above-mentioned simulation parameter decision method based on deep learning representation.

[0095] 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 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 need to comply with relevant regulations.

[0096] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. 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 methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0097] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0098] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered to be within the scope described in this specification.

[0099] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A simulation parameter decision-making method based on deep learning representation, characterized in that The simulation parameter decision-making method based on deep learning representation includes: Obtain the simulation data of historical vehicle design modeling, extract vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set; Apply the CFS method to screen out the key vehicle design parameters from the design parameter set to obtain a key design parameter set; the CFS method is a heuristic feature selection method based on statistical metrics; Use each key vehicle design parameter in the key design parameter set as the network input, and use the corresponding vehicle performance parameters in the performance parameter set as the network output to train the neural network KAN to obtain the trained neural network KAN; Derive a non-linear equation set between the key vehicle design parameters and each vehicle performance parameter according to the network structure of the trained neural network KAN; one non-linear equation in the non-linear equation set is a non-linear relationship formula between a vehicle performance parameter and each key vehicle design parameter; Take the satisfaction of each vehicle performance parameter with the set target as the parameter optimization target, and use the multi-objective optimization algorithm to solve the non-linear equation set to obtain the optimal vehicle design parameter value.

2. The simulation parameter decision method based on deep learning representation according to claim 1, wherein Obtain the simulation data of historical vehicle design modeling, extract vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set, specifically including: Obtain the simulation data of historical vehicle design modeling; Preprocess the simulation data to obtain the preprocessed simulation data; Normalize the preprocessed simulation data to obtain the normalized simulation data; Extract vehicle design parameters and vehicle performance parameters from the normalized simulation data to obtain a design parameter set and a performance parameter set.

3. The simulation parameter decision method based on deep learning representation according to claim 1, wherein Apply the CFS method to screen out the key vehicle design parameters from the design parameter set to obtain a key design parameter set, specifically including: Calculate the first correlation between the key vehicle design parameters and the vehicle performance parameters and the second correlation between the key vehicle design parameters; Calculate the Merit evaluation value of each vehicle design parameter in the design parameter set according to the first correlation and the second correlation; Screen out the vehicle design parameter with the largest Merit evaluation value as the key vehicle design parameter; add the screened key vehicle design parameter to the key design parameter set and calculate the Merit evaluation value of the current key design parameter set; initially, the key design parameter set is an empty set; Use the vehicle design parameter with the largest Merit evaluation value in the current remaining design parameter set and all the key vehicle design parameters in the current key design parameter set as a temporary subset, and calculate the Merit evaluation value of the temporary subset; the remaining design parameter set refers to the set composed of the parameters in the design parameter set except the current key vehicle design parameters; Judge whether the Merit evaluation value of the temporary subset is greater than the Merit evaluation value of the current key design parameter set; If so, use the temporary subset as the current set of key design parameters, and return to the step "Use the vehicle design parameter with the maximum Merit value in the current remaining design parameter set and all key vehicle design parameters in the current set of key design parameters as the temporary subset". If not, use the parameters in the current remaining design parameter set excluding the vehicle design parameters added to the temporary subset to form the current remaining design parameter set, and return to the step "Use the vehicle design parameter with the maximum Merit value in the current remaining design parameter set and all key vehicle design parameters in the current set of key design parameters as the temporary subset". Until the maximum value of the Merit value of the set of key design parameters is found to obtain the final set of key design parameters.

4. The simulation parameter decision-making method based on deep learning representation according to claim 3, characterized in that After performing the step "Apply the CFS method to screen out key vehicle design parameters from the design parameter set to obtain a set of key design parameters", the simulation parameter decision method based on deep learning representation further includes: evaluating the set of key design parameters, specifically: Divide the design parameter set into K subsets; Use each subset as the validation set, and the remaining K - 1 subsets as the training set; Use the training set to train the MLP model to obtain the first trained MLP model; Use the validation set to validate the first trained MLP model to obtain the first validation evaluation index; Determine the optimal first validation evaluation index; Use the set of key design parameters as the training set to train the MLP model to obtain the second trained MLP model; Use the remaining design parameter set as the validation set to validate the second trained MLP model to obtain the second validation evaluation index; Judge whether the second validation evaluation index is better than the optimal first validation evaluation index; If so, retain the set of key design parameters; If not, use the training set corresponding to the optimal first validation evaluation index as the set of key design parameters.

5. The simulation parameter decision method based on deep learning representation according to claim 1, wherein Taking the satisfaction of each vehicle performance parameter with the set target as the parameter optimization goal, use a multi-objective optimization algorithm to solve the non-linear equations to obtain the optimal vehicle design parameter values, specifically including: Taking the satisfaction of each vehicle performance parameter with the set target as the parameter optimization goal, determine the objective function; Use a genetic algorithm or a particle swarm optimization algorithm to solve the non-linear equations to obtain the optimal vehicle design parameter values; in the genetic algorithm or the particle swarm optimization algorithm, an individual in the population refers to a numerical combination of each key vehicle design parameter that satisfies the non-linear equations and the objective function.

6. A simulation parameter decision-making device based on deep learning representation, characterized in that, The simulation parameter decision device based on deep learning representation includes: A simulation data acquisition module, configured to acquire simulation data of historical vehicle design modeling, extract vehicle design parameters and vehicle performance parameters from the simulation data to obtain a design parameter set and a performance parameter set; A key vehicle design parameter screening module, configured to apply the CFS method to screen out key vehicle design parameters from the design parameter set to obtain a set of key design parameters; the CFS method is a heuristic feature selection method based on statistical metrics. The neural network KAN training module is used to train the neural network KAN with each key vehicle design parameter in the key design parameter set as the network input and each corresponding vehicle performance parameter in the performance parameter set as the network output, and obtain the trained neural network KAN; The non-linear equation system construction module is used to obtain the non-linear equation system between the key vehicle design parameters and each vehicle performance parameter according to the network structure of the trained neural network KAN; one non-linear equation in the non-linear equation system is a non-linear relationship formula between a vehicle performance parameter and each key vehicle design parameter; The solving module is used to take that each vehicle performance parameter meets the set target as the parameter optimization target, and use the multi-objective optimization algorithm to solve the non-linear equation system to obtain the optimal vehicle design parameter value.

7. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the simulation parameter decision method based on deep learning representation according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the simulation parameter decision method based on deep learning representation according to any one of claims 1-5.

Images