Machine learning based unmanned vehicle wheel-ground coupling model optimization method and system

Abstract

The application provides a kind of unmanned vehicle wheel ground coupling model optimization method and system based on machine learning, it is related to simulation technical field, the method comprises: by dividing model variable parameter and calibration parameter, construct simulation model, obtain simulation data set using Latin hypercube sampling, combined with preset driving condition to carry out real vehicle test and collect experimental data set;Gaussian process proxy model of simulation model is constructed to replace simulation model, Gaussian process model of modeling deviation function is built for modeling deviation, and data correlation is established;Integrate simulation data set and experimental data set, build joint Gaussian process model, calibration parameter calibration and model deviation correction are simultaneously completed by maximizing joint likelihood function, finally output response prediction value, the application does not need to rely on parameter prior distribution, distinguishes quantization modeling deviation and parameter deviation, greatly improves model precision and generalization ability, optimizes calculation efficiency.

Description

Technical Field

[0001] This invention relates to the field of simulation technology, specifically to a method and system for optimizing the wheel-ground coupling model of unmanned vehicles based on machine learning. Background Technology

[0002] The application of all-terrain unmanned vehicles in military reconnaissance, emergency rescue, geological exploration, and unmanned agriculture continues to deepen, making high-precision simulation of their mechanical behavior in interaction with off-road environments increasingly important. Wheel-ground coupling simulation models break through the traditional rigid ground assumption and can capture complex mechanical behaviors such as tire elastic deformation and soil compaction shear. However, existing models have two major problems: first, modeling deviations are caused by issues such as the rationality of assumptions, structural refinement, and simplification of boundary conditions during the modeling process; second, the limited means of parameter measurement and the dynamic evolution of terrain media make it difficult for core parameters to match the real scene, ultimately resulting in a large deviation between simulation output and actual mechanical behavior.

[0003] Traditional calibration methods heavily rely on prior distribution information of calibration parameters, forcibly incorporating them into standard models such as normal and uniform distributions. This disconnects them from the complex and ever-changing actual operating environment, resulting in calibration results that fail to reflect the true wheel-ground coupling mechanical behavior. Furthermore, traditional methods, due to their simplification assumptions, cannot distinguish between inherent biases in the quantification model and experimental observation errors, attributing both entirely to parameter deviations. Over-optimization of calibration parameters leads to calibration results deviating from physical properties, reducing the model's generalization ability. Therefore, a calibration method and system are urgently needed to address the insufficient predictive accuracy of existing models. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a machine learning-based method and system for optimizing unmanned vehicle-ground coupling models, thereby solving at least one of the problems in the background technology.

[0005] The first aspect of this invention is to provide a machine learning-based optimization method for an unmanned vehicle wheel-ground coupling model, the method comprising:

[0006] Acquire relevant data on the vehicle-to-ground coupling of all-terrain unmanned vehicles, divide them into model variable parameters and calibration parameters, and construct an all-terrain unmanned vehicle vehicle vehicle-to-ground coupling simulation model;

[0007] The Latin hypercube sampling method is used to sample within the value range of model variable parameters and calibration parameters to drive batch simulation of the simulation model and obtain simulation dataset. At the same time, sampling is also performed within the value range of model variable parameters, and real vehicle tests are carried out in combination with preset driving conditions to collect observation data and form experimental dataset.

[0008] Based on the simulation dataset, a Gaussian process surrogate model of the simulation model is constructed, and the hyperparameters of the Gaussian process surrogate model are estimated by the maximum likelihood estimation method. The simulation response prediction values ​​of the unsampled points are then output.

[0009] Based on the modeling deviation between the experimental dataset and the predicted simulation response, a Gaussian process model of the modeling deviation function is constructed, and a correlation is established between the Gaussian process surrogate model and the experimental error parameters.

[0010] By integrating the simulation dataset and the experimental dataset, a joint Gaussian process model is constructed.

[0011] Based on the joint Gaussian process model, a joint likelihood function is constructed. The joint likelihood function is maximized through an optimization algorithm to solve for the optimal calibration parameters, hyperparameters of the Gaussian process model, and correlation coefficients of the modeling bias function.

[0012] The optimal calibration parameters are used as the calibrated calibration parameters and input together with the model variable parameters to be predicted into the joint Gaussian process model to obtain the calibrated response prediction value.

[0013] According to one aspect of the above technical solution, the steps of acquiring all-terrain unmanned vehicle wheel-ground coupling related data, dividing it into model variable parameters and calibration parameters, and constructing an all-terrain unmanned vehicle wheel-ground coupling simulation model specifically include:

[0014] Based on the all-terrain unmanned vehicle wheel-ground coupling correlation data, the model variable parameters are divided into experimentally observable model variable parameters and non-experimentally observable calibration parameters. The calibration parameters include at least one of soil shear modulus, soil cohesion, wheel-soil contact friction coefficient, soil interparticle restitution coefficient, and soil interparticle static friction coefficient. The model variable parameters include at least one of wheel structure dimensions, vehicle mass, suspension system parameters, and driving speed.

[0015] The range of values ​​for calibration parameters is set based on expert information or engineering experience.

[0016] A wheel-to-ground contact simulation model was constructed using the discrete element method, and an unmanned vehicle dynamics model was built using multibody dynamics simulation software to establish an all-terrain unmanned vehicle wheel-to-ground coupling simulation model.

[0017] According to one aspect of the above technical solution, based on the simulation dataset, the steps of constructing a Gaussian process surrogate model for the simulation model, estimating the hyperparameters of the Gaussian process surrogate model using the maximum likelihood estimation method, and outputting the predicted simulation response values ​​for unsampled points specifically include:

[0018] Based on the simulation dataset, a Gaussian process surrogate model for the simulation model is constructed, represented as:

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] ,

[0024] ,

[0025] in, The model is a Gaussian process. The output response value of the simulation dataset, The model variable parameters input to the simulation dataset, The calibration parameters are input to the simulation dataset. Let be the mean function of the Gaussian process surrogate model. This is a matrix consisting of the known regression term vectors of the Gaussian process surrogate model. These are the hyperparameters to be estimated in the Gaussian process surrogate model. For simulation datasets , The covariance of the two inputs, The variance to be estimated is... For the correlation matrix, For the first The hyperparameters that need to be estimated are the correlations of the variables. The set of hyperparameters whose correlation needs to be estimated. , The first Group, Group simulation dataset, for , The correlation between them The number of groups in the simulation dataset. , The first Group, The first in the group simulation dataset One variable, The total number of variables;

[0026] The hyperparameters of the Gaussian process surrogate model are estimated using the maximum likelihood estimation method, and are expressed as follows:

[0027] ,

[0028] ,

[0029] in, The term "hyperparameters" refers to the collective hyperparameters to be estimated in the Gaussian process surrogate model of the simulation model. , , , for The likelihood function.

[0030] According to one aspect of the above technical solution, the step of constructing a Gaussian process model of the modeling deviation function based on the modeling deviation between the experimental dataset and the simulated response prediction value, and establishing a correlation between the Gaussian process surrogate model and the experimental error parameters, specifically includes:

[0031] Based on the modeling deviation between the experimental dataset and the predicted simulation response, a Gaussian process model of the modeling deviation function is constructed, expressed as:

[0032] ,

[0033] in, To model a Gaussian process with a bias function, Model variable parameters in the experimental dataset Modeling deviation function, The mean function of the Gaussian process model for modeling the deviation function. Covariance for modeling the Gaussian process model of the bias function;

[0034] The correlation between the Gaussian process surrogate model and experimental error parameters is established and expressed as follows:

[0035] ,

[0036] in, Model variable parameters The output response value, from or Take the value from the middle. The correlation coefficient has been estimated. The predicted simulation output response value. The predicted modeling bias value. This represents experimental error.

[0037] According to one aspect of the above technical solution, the step of integrating the simulation dataset and the experimental dataset to construct a joint Gaussian process model specifically includes:

[0038] By integrating the simulation dataset and the experimental dataset, a joint dataset is obtained, represented as:

[0039] ,

[0040] in, For a joint dataset, The input parameter space and output response of the joint dataset;

[0041] The Gaussian process model based on the Gaussian process surrogate model and the modeling deviation function is expressed as follows:

[0042] ,

[0043] ,

[0044] ,

[0045] ,

[0046] in, The mean function of the joint Gaussian process model. This is a matrix composed of the known regression term vectors of the joint Gaussian process model. , The matrices are composed of the known regression term vectors of the Gaussian process surrogate model for the simulation model and the Gaussian process model for the modeling bias function, respectively. The hyperparameters to be estimated in the joint Gaussian process model are... For Gaussian process proxy model , The covariance of the two inputs, The calibration parameters to be calibrated For the covariance of the joint Gaussian process model, , These are the experimental datasets. , Two sets of input combinations of calibration parameters to be calibrated As input to the Gaussian process surrogate model, for × The identity matrix, The standard deviation of the experimental error. For the covariance of the joint Gaussian process model, The correlation coefficient is to be estimated.

[0047] According to one aspect of the above technical solution, the steps of constructing a joint likelihood function based on the joint Gaussian process model, maximizing the joint likelihood function through an optimization algorithm, and solving for the optimal calibration parameters, modeling bias function, hyperparameters, and correlation coefficients of the Gaussian process model, specifically include:

[0048] Based on the joint Gaussian process model, a joint likelihood function is constructed. This joint likelihood function is then maximized using an optimization algorithm, and expressed as:

[0049] ,

[0050] ,

[0051] in, For the joint likelihood function, The number of groups in the experimental dataset. The hyperparameters to be estimated for the Gaussian process model that models the deviation function are... This refers to the collective hyperparameters estimated in the Gaussian process surrogate model.

[0052] According to one aspect of the above technical solution, the optimal calibration parameters are used as the calibrated calibration parameters, and input together with the model variable parameters to be predicted into the joint Gaussian process model to obtain the calibrated response prediction value, expressed as:

[0053] ,

[0054] in, These are the calibrated standard parameters. , , They are respectively the estimated , , , This is the calibrated predicted response value.

[0055] A second aspect of the present invention provides a machine learning-based unmanned vehicle wheel-ground coupling model optimization system, the system being used to implement the above-mentioned machine learning-based unmanned vehicle wheel-ground coupling model optimization method, the system comprising:

[0056] The data acquisition module is used to acquire all-terrain unmanned vehicle wheel-ground coupling related data, divide it into model variable parameters and calibration parameters, and construct an all-terrain unmanned vehicle wheel-ground coupling simulation model;

[0057] The data processing module is used to sample within the value range of model variable parameters and calibration parameters using the Latin hypercube sampling method, drive the simulation model to perform batch simulations, obtain simulation datasets, and simultaneously sample within the value range of model variable parameters, conduct real vehicle tests in combination with preset driving conditions, collect observation data, and form experimental datasets.

[0058] The Gaussian process surrogate model construction module is used to construct a Gaussian process surrogate model for the simulation model based on the simulation dataset, estimate the hyperparameters of the Gaussian process surrogate model using the maximum likelihood estimation method, and output the simulation response prediction values ​​of the unsampled points.

[0059] The correlation construction module is used to construct a Gaussian process model of the modeling deviation function based on the modeling deviation between the experimental dataset and the simulation response prediction value, and to establish a correlation relationship by combining the Gaussian process surrogate model and experimental error parameters.

[0060] The joint Gaussian process model construction module is used to integrate the simulation dataset and the experimental dataset to construct a joint Gaussian process model;

[0061] The model optimization module is used to construct a joint likelihood function based on the joint Gaussian process model, maximize the joint likelihood function through an optimization algorithm, and solve for the optimal calibration parameters, modeling bias function, hyperparameters, and correlation coefficients of the Gaussian process model.

[0062] The prediction output module is used to take the optimal calibration parameters as the calibrated calibration parameters, and input them together with the model variable parameters to be predicted into the Gaussian process model to obtain the calibrated response prediction value.

[0063] A third aspect of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described machine learning-based unmanned vehicle-ground coupling model optimization method.

[0064] A fourth aspect of the present invention relates to an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the aforementioned machine learning-based unmanned vehicle-ground coupling model optimization method.

[0065] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0066] 1. By constructing a Gaussian process model for the modeling deviation function separately, the inherent deviation of the model and the parameter deviation can be quantitatively distinguished. This avoids the over-optimization problem caused by traditional methods that attribute all errors to parameter deviation, ensuring that the calibration parameters always conform to the physical reality. At the same time, by leveraging the accurate nonlinear fitting capability of the Gaussian process surrogate model to the original wheel-ground coupling simulation model, reliable data support and computational foundation are provided for the quantitative distinction of deviations. While ensuring the accuracy of the distinction, the overall computational efficiency is greatly improved, and the time cost of batch simulation and calibration is significantly reduced.

[0067] 2. Breaking away from the strong dependence of traditional calibration methods on prior distribution information of parameters, it eliminates the need to forcibly incorporate calibration parameters into standard models such as normal or uniform distributions, making it more suitable for complex and ever-changing actual operating environments. By combining the maximum likelihood estimation hyperparameter solution method of Gaussian process surrogate model, it simultaneously completes calibration parameter calibration and model bias correction, making the calibration parameters more consistent with the wheel-ground coupling mechanical behavior of real scenarios, ultimately significantly improving the model prediction accuracy. Attached Figure Description

[0068] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0069] Figure 1 This is a schematic diagram of the all-terrain unmanned vehicle wheel-ground coupling simulation model in Embodiment 1 of the present invention;

[0070] Figure 2 This is a diagram showing the output prediction results of the all-terrain unmanned vehicle wheel-ground coupling simulation model after calibration in Embodiment 1 of the present invention;

[0071] Figure 3 This is a graph showing the absolute error of the output prediction after calibration of the all-terrain unmanned vehicle wheel-ground coupling simulation model in Embodiment 1 of the present invention. Detailed Implementation

[0072] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be more thorough and complete.

[0073] Example 1

[0074] Embodiment 1 of the present invention provides an optimization method for an unmanned vehicle wheel-ground coupling model based on machine learning, the method comprising steps S1-S7:

[0075] Step S1: Obtain relevant data on the vehicle-to-ground coupling of all-terrain unmanned vehicles, divide them into model variable parameters and calibration parameters, and construct an all-terrain unmanned vehicle-to-ground coupling simulation model.

[0076] Specifically, based on the all-terrain unmanned vehicle wheel-ground coupling related data, the model variable parameters are divided into experimentally observable model variable parameters and non-experimentally observable calibration parameters. The calibration parameters include at least one of soil shear modulus, soil cohesion, wheel-soil contact friction coefficient, soil particle restitution coefficient, and soil particle static friction coefficient. The model variable parameters include at least one of wheel structure size, vehicle mass, suspension system parameters, and driving speed.

[0077] The range of values ​​for calibration parameters is set based on expert information or engineering experience.

[0078] A wheel-ground contact simulation model was constructed using the Discrete Element Method (DEM), and an unmanned vehicle dynamics model was constructed using Multibody Dynamics (MBD) simulation software. A DEM-MBD joint all-terrain unmanned vehicle wheel-ground coupling simulation model was then established.

[0079] As an example and not a limitation, the model variable parameters selected in this embodiment are the suspension spring stiffness and suspension damping coefficient of the all-terrain unmanned vehicle, and the coefficient of restitution and static friction between soil particles in the EDEM software are selected as calibration parameters. The corresponding distribution range is queried from the general material model database in EDEM based on the intrinsic soil information.

[0080] Furthermore, some parameters in the all-terrain wheel-ground coupling simulation model have a slight impact on wheel-ground coupling characteristics. These parameters do not need to be included in the calibration scope. Only the parameters that have a significant impact on wheel-ground coupling characteristics are selected for calibration. This can not only avoid the risk of overfitting caused by parameter redundancy in full calibration and ensure the model's generalization ability under multiple working conditions, but also reduce data requirements, lower acquisition costs, simplify the process, and improve engineering feasibility while ensuring accuracy.

[0081] Please see Figure 1 As shown in this embodiment, when building the all-terrain unmanned vehicle wheel-ground coupled simulation model, a discrete element model of the soil road surface is established using EDEM software, and a multibody dynamics model of the all-terrain unmanned vehicle is established using Motionview software. The two can achieve DEM-MBD joint simulation through the coupling interface to accurately simulate the wheel-ground interaction and the dynamic response of the whole vehicle.

[0082] Step S2: The Latin hypercube sampling method is used to sample within the value range of model variable parameters and calibration parameters to drive the simulation model to perform batch simulations and obtain a simulation dataset. At the same time, sampling is performed within the value range of model variable parameters, and real vehicle tests are carried out in combination with preset driving conditions to collect observation data and form an experimental dataset.

[0083] Examples, rather than limitations, integrate model variable parameters and calibration parameters to form a 4- to 6-dimensional input parameter space (e.g., 2 model variable parameters + 2 calibration parameters).

[0084] For example, Latin hypercube sampling was used to generate 300 sets of simulation sample points. The working conditions were set as soft soil horizontal road surface, initial speed of 1m / s, and driving for 5s (the first 4s were for transition, and the last 1s was for steady state). The average value of the vehicle body center of gravity drop height during the steady state stage was collected as the output response to form a simulation dataset.

[0085] Similarly, for example, 20 sets of samples are taken within the range of model variable parameters, and real vehicle tests are carried out in combination with preset driving conditions (such as soft soil level road surface, initial speed of 1m / s, driving for 5s) to collect observation data and obtain experimental datasets.

[0086] Step S3: Based on the simulation dataset, construct a Gaussian process surrogate model for the simulation model, estimate the hyperparameters of the Gaussian process surrogate model using the maximum likelihood estimation method, and output the simulation response prediction values ​​for unsampled points.

[0087] Specifically, based on the simulation dataset, a Gaussian process surrogate model for the simulation model is constructed. The Gaussian process surrogate model is a Gaussian process model. Because it has the advantages of strong nonlinear fitting ability, prediction results with confidence intervals, and no need for training with a large amount of data, it is suitable to replace the time-consuming all-terrain unmanned vehicle wheel-ground coupling simulation model.

[0088] The prior of the Gaussian process surrogate model can be defined by the mean function and covariance, and the correlation of the data is described using a Gaussian correlation function, expressed as:

[0089] ,

[0090] ,

[0091] ,

[0092] ,

[0093] ,

[0094] ,

[0095] in, The model is a Gaussian process. The output response value of the simulation dataset, The model variable parameters input to the simulation dataset, The calibration parameters are input to the simulation dataset. Let be the mean function of the Gaussian process surrogate model. This is a matrix consisting of the known regression term vectors of the Gaussian process surrogate model. These are the hyperparameters to be estimated in the Gaussian process surrogate model. For simulation datasets , The covariance of the two inputs, The variance to be estimated is... The correlation matrix is ​​calculated using a Gaussian correlation function. For the first The hyperparameters that need to be estimated are the correlations of the variables. The set of hyperparameters whose correlation needs to be estimated. , The first Group, Group simulation dataset, for , The correlation between them The number of groups in the simulation dataset. , The first Group, The first in the group simulation dataset One variable, This represents the total number of variables.

[0096] Understandable. It contains two types of parameter groups: one is the model variable parameters. (Composed of multiple model variables and parameter variables, such as suspension damping coefficient, spring height, wheel structure dimensions, etc. in practical applications), and secondly, calibration parameters. (Composed of multiple calibration parameter variables, such as the soil particle restitution coefficient, soil-wheel contact static friction coefficient, and soil cohesion in actual scenarios). Similarly, in the subsequent Gaussian process model of the bias function, there is also a correlation matrix, but the parameters involved here are the model variable parameters of the experimental dataset. (correspond (The multiple model parameter variables included).

[0097] The hyperparameters of the Gaussian process surrogate model are estimated using the maximum likelihood estimation method, and are expressed as follows:

[0098] ,

[0099] ,

[0100] in, The term "hyperparameters" refers to the collective hyperparameters to be estimated in the Gaussian process surrogate model of the simulation model. , , , for The likelihood function, This represents the number of groups in the simulation dataset.

[0101] Furthermore, an evolutionary algorithm is used for iterative solution: the likelihood function value corresponding to the Gaussian process surrogate model is used as the optimization objective. In each iteration, the likelihood function value corresponding to the current parameter set is calculated. The preset number of evaluations of the Gaussian process surrogate model is used as the convergence condition for the iteration. When the number of iterations reaches the preset value, the set of parameters with the largest likelihood function value among all iterations is selected as the final solution output.

[0102] Step S4: Based on the modeling deviation between the experimental dataset and the predicted simulation response, construct a Gaussian process model of the modeling deviation function, and establish a correlation between the Gaussian process surrogate model and the experimental error parameters.

[0103] Specifically, the prior of the Gaussian process model can be defined by the mean function and covariance, expressed as:

[0104] ,

[0105] in, To model a Gaussian process with a bias function, Model variable parameters in the experimental dataset Modeling deviation function, The mean function of the Gaussian process model for modeling the deviation function. The covariance of the Gaussian process model for modeling the bias function is given. Similarly, the Gaussian process model for modeling the bias function also includes the total number of relevant hyperparameters to be estimated. ;

[0106] Similarly, a Gaussian correlation function is used to describe the data correlation, and the output is the inherent bias value implied by the Gaussian process surrogate model under the corresponding working condition. The Gaussian process model constructed for it also includes the correlation hyperparameters to be estimated. .

[0107] The correlation between the Gaussian process surrogate model and experimental error parameters is established and expressed as follows:

[0108] ,

[0109] in, Model variable parameters The output response value, from or Take the value from the middle. The correlation coefficient has been estimated. The predicted simulation output response value. The predicted modeling bias value. This is experimental error;

[0110] in, To account for experimental error, it is generally assumed that the distribution follows a zero-mean normal distribution, i.e. λ is the standard deviation of the experimental error, which is set according to the preset working conditions.

[0111] Step S5: Integrate the simulation dataset and the experimental dataset to construct a joint Gaussian process model;

[0112] Specifically, the simulation dataset and the experimental dataset are integrated to obtain a joint dataset, represented as:

[0113] ,

[0114] in, For a joint dataset, The input parameter space and output response of the joint dataset;

[0115] Based on the Gaussian process surrogate model and the Gaussian process model with the modeling deviation function, a joint Gaussian process model is constructed by fusing the mean function and the covariance function. The Gaussian process model is still used, fully utilizing the correlation information between simulation data and experimental data, and is expressed as follows:

[0116] ,

[0117] ,

[0118] ,

[0119] ,

[0120] in, The mean function of the joint Gaussian process model. This is a matrix composed of the known regression term vectors of the joint Gaussian process model. , The matrices are composed of the known regression term vectors of the Gaussian process surrogate model for the simulation model and the Gaussian process model for the modeling bias function, respectively. The hyperparameters to be estimated in the joint Gaussian process model are... For Gaussian process proxy model , The covariance of the two inputs, The calibration parameters to be calibrated For the covariance of the joint Gaussian process model, , These are the experimental datasets. , Two sets of input combinations of calibration parameters to be calibrated As input to the Gaussian process surrogate model, for × The identity matrix, The standard deviation of the experimental error. For the covariance of the joint Gaussian process model, The correlation coefficient is to be estimated.

[0121] Step S6: Construct a joint likelihood function based on the joint Gaussian process model, maximize the joint likelihood function through an optimization algorithm, and solve for the optimal calibration parameters, hyperparameters of the Gaussian process model of the modeling bias function, and correlation coefficients.

[0122] Specifically, a joint likelihood function is constructed based on the joint Gaussian process model, and the joint likelihood function is maximized through an optimization algorithm, expressed as:

[0123] ,

[0124] ,

[0125] in, For the joint likelihood function, The number of groups in the experimental dataset. The hyperparameters to be estimated for the Gaussian process model that models the deviation function are... This refers to the collective hyperparameters estimated in the Gaussian process surrogate model.

[0126] Furthermore, an evolutionary algorithm is used for iterative solution: taking the joint likelihood function corresponding to the joint Gaussian process model as the optimization objective, the joint likelihood function value corresponding to the current parameter set is calculated in each iteration; the preset number of evaluations of the joint Gaussian process model is used as the iteration convergence condition. When the number of iterations reaches the preset value, the set of parameters with the largest joint likelihood function value among all iterations is selected as the final solution output.

[0127] The optimal calibration parameters, hyperparameters of the Gaussian process model of the modeling deviation function, and correlation coefficients are obtained by solving the problem. Simultaneously, the calibration parameters are calibrated and the inherent deviations of the simulation model are corrected. In other words, the inherent deviations caused by the simplification assumptions made when the all-terrain unmanned vehicle wheel-ground coupling simulation model simulates the real physical process are corrected.

[0128] Step S7: The optimal calibration parameters are used as the calibrated calibration parameters and input together with the model variable parameters to be predicted into the joint Gaussian process model to obtain the calibrated response prediction value.

[0129] Represented as:

[0130] ,

[0131] in, These are the calibrated standard parameters. , , They are respectively the estimated , , , This is the calibrated predicted response value.

[0132] according to Figures 2-3 It can be seen that the predicted surface and the actual response surface almost completely overlap. Even at the edge of the input parameter combination, i.e. the extreme working condition area, the deviation between the two can be ignored, which shows that the prediction accuracy of the all-terrain unmanned wheel-ground coupling simulation model after optimization by this optimization method is high.

[0133] To further illustrate this embodiment, the root mean square error (RMSE) of the optimized prediction is used as the evaluation metric to measure the prediction accuracy after model calibration.

[0134] Table 1:

[0135]

[0136] The traditional model calibration method performs steps S1-S3. The subsequent steps are: statistical inference of the experimental dataset using the Bayesian algorithm, updating the preset prior distribution (the range of values ​​for calibration parameters set according to expert information or engineering experience information) to the posterior distribution (the range of values ​​for inferred calibration parameters) using the likelihood function, and then performing model response prediction based on the posterior distribution.

[0137] As shown in Table 1, the method of this embodiment has high prediction accuracy, which demonstrates the obvious advantages of the optimization method of this embodiment. Therefore, the method of this embodiment performs well in terms of prediction accuracy of the all-terrain unmanned vehicle wheel-ground coupling simulation model.

[0138] Example 2

[0139] Embodiment 2 of the present invention provides an optimization system for an unmanned vehicle wheel-ground coupling model based on machine learning, the system comprising:

[0140] The data acquisition module is used to acquire all-terrain unmanned vehicle wheel-ground coupling related data, divide it into model variable parameters and calibration parameters, and construct an all-terrain unmanned vehicle wheel-ground coupling simulation model;

[0141] The data processing module is used to sample within the value range of model variable parameters and calibration parameters using the Latin hypercube sampling method, drive the simulation model to perform batch simulations, obtain simulation datasets, and simultaneously sample within the value range of model variable parameters, conduct real vehicle tests in combination with preset driving conditions, collect observation data, and form experimental datasets.

[0142] The Gaussian process surrogate model construction module is used to construct a Gaussian process surrogate model for the simulation model based on the simulation dataset, estimate the hyperparameters of the Gaussian process surrogate model using the maximum likelihood estimation method, and output the simulation response prediction values ​​of the unsampled points.

[0143] The correlation construction module is used to construct a Gaussian process model of the modeling deviation function based on the modeling deviation between the experimental dataset and the simulation response prediction value, and to establish a correlation relationship by combining the Gaussian process surrogate model and experimental error parameters.

[0144] The joint Gaussian process model construction module is used to integrate the simulation dataset and the experimental dataset to construct a joint Gaussian process model;

[0145] The model optimization module is used to construct a joint likelihood function based on the joint Gaussian process model, maximize the joint likelihood function through an optimization algorithm, and solve for the optimal calibration parameters, modeling bias function, hyperparameters, and correlation coefficients of the Gaussian process model.

[0146] The prediction output module is used to take the optimal calibration parameters as the calibrated calibration parameters, and input them together with the model variable parameters to be predicted into the Gaussian process model to obtain the calibrated response prediction value.

[0147] Example 3

[0148] Embodiment 3 of the present invention provides a readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method described in Embodiment 1.

[0149] Example 4

[0150] Embodiment 4 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described in Embodiment 1.

[0151] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0152] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0153] More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable storage media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0154] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0155] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0156] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A machine learning-based optimization method for an unmanned vehicle wheel-ground coupling model, characterized in that, The method includes: Acquire relevant data on the vehicle-to-ground coupling of all-terrain unmanned vehicles, divide them into model variable parameters and calibration parameters, and construct an all-terrain unmanned vehicle vehicle vehicle-to-ground coupling simulation model; The Latin hypercube sampling method is used to sample within the value range of model variable parameters and calibration parameters to drive batch simulation of the simulation model and obtain simulation dataset. At the same time, sampling is also performed within the value range of model variable parameters, and real vehicle tests are carried out in combination with preset driving conditions to collect observation data and form experimental dataset. Based on the simulation dataset, a Gaussian process surrogate model of the simulation model is constructed, and the hyperparameters of the Gaussian process surrogate model are estimated by the maximum likelihood estimation method. The simulation response prediction values ​​of the unsampled points are then output. Based on the modeling deviation between the experimental dataset and the predicted simulation response, a Gaussian process model of the modeling deviation function is constructed, and a correlation is established between the Gaussian process surrogate model and the experimental error parameters. By integrating the simulation dataset and the experimental dataset, a joint Gaussian process model is constructed. Based on the joint Gaussian process model, a joint likelihood function is constructed. The joint likelihood function is maximized using an optimization algorithm to solve for the optimal calibration parameters, the hyperparameters of the Gaussian process model, and the correlation coefficients of the modeling bias function, including: Based on the joint Gaussian process model, a joint likelihood function is constructed. This joint likelihood function is then maximized using an optimization algorithm, and expressed as: ， ， in, For the joint likelihood function, The number of groups in the experimental dataset. The hyperparameters to be estimated for the Gaussian process model that models the deviation function are... This refers to the collective hyperparameters estimated in the Gaussian process surrogate model. For a joint dataset, For the output response of the joint dataset, The calibration parameters to be calibrated The correlation coefficient to be estimated is... The number of groups in the simulation dataset. For the covariance of the joint Gaussian process model, This is a matrix composed of the known regression term vectors of the joint Gaussian process model. These are the hyperparameters to be estimated for the joint Gaussian process model; The optimal calibration parameters are used as the calibrated parameters and input along with the model variable parameters to be predicted into the joint Gaussian process model to obtain the calibrated response prediction value, expressed as: ， in, These are the calibrated standard parameters. , , They are respectively the estimated , , , The calibrated response prediction value. Model variable parameters The output response value.

2. The machine learning-based unmanned vehicle-ground coupling model optimization method according to claim 1, characterized in that, The steps for acquiring all-terrain unmanned vehicle wheel-ground coupling related data, dividing it into model variable parameters and calibration parameters, and constructing an all-terrain unmanned vehicle wheel-ground coupling simulation model specifically include: Based on the all-terrain unmanned vehicle wheel-ground coupling correlation data, the model variable parameters are divided into experimentally observable model variable parameters and non-experimentally observable calibration parameters. The calibration parameters include at least one of soil shear modulus, soil cohesion, wheel-soil contact friction coefficient, soil interparticle restitution coefficient, and soil interparticle static friction coefficient. The model variable parameters include at least one of wheel structure dimensions, vehicle mass, suspension system parameters, and driving speed. The range of values ​​for calibration parameters is set based on expert information or engineering experience. A wheel-to-ground contact simulation model was constructed using the discrete element method, and an unmanned vehicle dynamics model was built using multibody dynamics simulation software to establish an all-terrain unmanned vehicle wheel-to-ground coupling simulation model.

3. The machine learning-based unmanned vehicle-ground coupling model optimization method according to claim 1, characterized in that, Based on the simulation dataset, the steps of constructing a Gaussian process surrogate model for the simulation model, estimating the hyperparameters of the Gaussian process surrogate model using the maximum likelihood estimation method, and outputting the predicted simulation response values ​​for unsampled points specifically include: Based on the simulation dataset, a Gaussian process surrogate model for the simulation model is constructed, represented as: ， ， ， ， ， ， in, For Gaussian process model, The output response value of the simulation dataset, The model variable parameters input to the simulation dataset, The calibration parameters are input to the simulation dataset. Let be the mean function of the Gaussian process surrogate model. This is a matrix consisting of the known regression term vectors of the Gaussian process surrogate model. These are the hyperparameters to be estimated in the Gaussian process surrogate model. For simulation datasets , The covariance of the two inputs, The variance to be estimated is... For the correlation matrix, For the first The hyperparameters that need to be estimated are the correlations of the variables. The set of hyperparameters whose correlation needs to be estimated. , The first Group, Group simulation dataset, for , The correlation between them , The first Group, The first in the group simulation dataset One variable, The total number of variables; The hyperparameters of the Gaussian process surrogate model are estimated using the maximum likelihood estimation method, and are expressed as follows: ， ， in, The term "hyperparameters" refers to the collective hyperparameters to be estimated in the Gaussian process surrogate model of the simulation model. , , , for The likelihood function.

4. The machine learning-based unmanned vehicle-ground coupling model optimization method according to claim 3, characterized in that, Based on the modeling deviation between the experimental dataset and the predicted simulation response, a Gaussian process model of the modeling deviation function is constructed, and a correlation is established between the Gaussian process surrogate model and the experimental error parameters. Specifically, this includes: Based on the modeling deviation between the experimental dataset and the predicted simulation response, a Gaussian process model of the modeling deviation function is constructed, expressed as: ， in, To model a Gaussian process with a bias function, Model variable parameters in the experimental dataset Modeling deviation function, The mean function of the Gaussian process model for modeling the deviation function. Covariance for modeling the Gaussian process model of the bias function; The correlation between the Gaussian process surrogate model and experimental error parameters is established and expressed as follows: ， in, from or Take the value from the middle. The correlation coefficient has been estimated. The predicted simulation output response value, The predicted modeling bias value. This represents experimental error.

5. The machine learning-based unmanned vehicle-ground coupling model optimization method according to claim 4, characterized in that, The steps for integrating the simulation dataset and the experimental dataset to construct a joint Gaussian process model specifically include: By integrating the simulation dataset and the experimental dataset, a joint dataset is obtained, represented as: ， in, For a joint dataset, The input parameter space and output response of the joint dataset; The Gaussian process model based on the Gaussian process surrogate model and the modeling deviation function is expressed as follows: ， ， ， ， in, The mean function of the joint Gaussian process model. , The matrices are composed of the known regression term vectors of the Gaussian process surrogate model for the simulation model and the Gaussian process model for the modeling bias function, respectively. For Gaussian process proxy model , The covariance of the two inputs, , These are the experimental datasets. , Two sets of input combinations of calibration parameters to be calibrated As input to the Gaussian process surrogate model, for × The identity matrix, The standard deviation of the experimental error. This represents the covariance of the joint Gaussian process model.

6. A machine learning-based unmanned vehicle wheel-ground coupling model optimization system, characterized in that, The system is used to implement the machine learning-based unmanned vehicle-ground coupling model optimization method according to any one of claims 1 to 5, and the system comprises: The data acquisition module is used to acquire all-terrain unmanned vehicle wheel-ground coupling related data, divide it into model variable parameters and calibration parameters, and construct an all-terrain unmanned vehicle wheel-ground coupling simulation model; The data processing module is used to sample within the value range of model variable parameters and calibration parameters using the Latin hypercube sampling method, drive the simulation model to perform batch simulations, obtain simulation datasets, and simultaneously sample within the value range of model variable parameters, conduct real vehicle tests in combination with preset driving conditions, collect observation data, and form experimental datasets. The Gaussian process surrogate model construction module is used to construct a Gaussian process surrogate model for the simulation model based on the simulation dataset, estimate the hyperparameters of the Gaussian process surrogate model using the maximum likelihood estimation method, and output the simulation response prediction values ​​of the unsampled points. The correlation construction module is used to construct a Gaussian process model of the modeling deviation function based on the modeling deviation between the experimental dataset and the simulation response prediction value, and to establish a correlation relationship by combining the Gaussian process surrogate model and experimental error parameters. The joint Gaussian process model construction module is used to integrate the simulation dataset and the experimental dataset to construct a joint Gaussian process model; The model optimization module is used to construct a joint likelihood function based on the joint Gaussian process model, maximize the joint likelihood function through an optimization algorithm, and solve for the optimal calibration parameters, modeling bias function, hyperparameters, and correlation coefficients of the Gaussian process model, including: Based on the joint Gaussian process model, a joint likelihood function is constructed. This joint likelihood function is then maximized using an optimization algorithm, and expressed as: ， ， in, For the joint likelihood function, The number of groups in the experimental dataset. The hyperparameters to be estimated for the Gaussian process model that models the deviation function are... This refers to the collective hyperparameters estimated in the Gaussian process surrogate model. For a joint dataset, For the output response of the joint dataset, The calibration parameters to be calibrated The correlation coefficient to be estimated is... The number of groups in the simulation dataset. For the covariance of the joint Gaussian process model, This is a matrix composed of the known regression term vectors of the joint Gaussian process model. These are the hyperparameters to be estimated for the joint Gaussian process model; The prediction output module takes the optimal calibration parameters as the calibrated calibration parameters and inputs them, along with the model variable parameters to be predicted, into the joint Gaussian process model to obtain the calibrated response prediction value, expressed as: ， in, These are the calibrated standard parameters. , , They are respectively the estimated , , , The calibrated response prediction value. Model variable parameters The output response value.

7. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the machine learning-based unmanned vehicle-ground coupling model optimization method as described in any one of claims 1 to 5.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the machine learning-based unmanned wheel-ground coupling model optimization method as described in any one of claims 1 to 5.

Images