CN-121835453-B - Unmanned wheel ground coupling model optimization method and system based on machine learning

CN121835453BCN 121835453 BCN121835453 BCN 121835453BCN-121835453-B

Abstract

The invention provides an unmanned wheel ground coupling model optimization method and system based on machine learning, and relates to the technical field of simulation, wherein the method comprises the steps of constructing a simulation model by dividing model variable parameters and calibration parameters, acquiring a simulation data set by Latin hypercube sampling, and developing a real vehicle test acquisition experimental data set by combining preset driving conditions; the invention discloses a Gaussian process proxy model, which comprises the steps of constructing a Gaussian process proxy model of a simulation model, replacing the simulation model, constructing a Gaussian process model of a modeling deviation function aiming at modeling deviation, establishing data association, integrating a simulation data set and an experimental data set, constructing a joint Gaussian process model, synchronously completing calibration parameter calibration and model deviation correction by maximizing a joint likelihood function, and finally outputting a response predicted value.

Inventors

JIANG CHEN
CHEN JIALE
QIU HAOBO
MENG LEI
MA RUI

Assignees

华中科技大学

Dates

Publication Date: 20260512
Application Date: 20260316

Claims (8)

1. A machine learning-based unmanned wheel ground coupling model optimization method, the method comprising: Acquiring all-terrain unmanned wheel ground coupling related data, dividing the data into model variable parameters and calibration parameters, and constructing an all-terrain unmanned wheel ground coupling simulation model; Sampling in a value interval of model variable parameters and calibration parameters by using a Latin hypercube sampling method, driving a simulation model to perform batch simulation to obtain a simulation data set, sampling in the value interval of the model variable parameters, carrying out a real vehicle test in combination with a preset driving working condition, and collecting observation data to form an experimental data set; Based on the simulation data set, constructing a Gaussian process proxy model of a simulation model, estimating super parameters of the Gaussian process proxy model by adopting a maximum likelihood estimation method, and outputting a simulation response predicted value of an un-sampled point; Based on modeling deviation between the experimental data set and the simulation response predicted value, constructing a Gaussian process model of a modeling deviation function, and establishing an association relationship by combining a Gaussian process proxy model and experimental error parameters; Integrating the simulation data set and the experimental data set to construct a combined Gaussian process model; Constructing a joint likelihood function based on the joint Gaussian process model, maximizing the joint likelihood function through an optimization algorithm, and solving optimal calibration parameters, super parameters and correlation coefficients of the Gaussian process model modeling a deviation function, wherein the method comprises the following steps: Constructing a joint likelihood function based on the joint Gaussian process model, and maximizing the joint likelihood function through an optimization algorithm, wherein the joint likelihood function is expressed as: , , Wherein, the In order to combine the likelihood functions, For the number of sets of experimental data sets, To model the hyper-parameters to be estimated of the gaussian process model of the deviation function, To summarize the estimated hyper-parameters in the gaussian process proxy model, In order to combine the data sets, In order to combine the output responses of the data sets, As the calibration parameter to be calibrated, For the correlation coefficient to be estimated, In order to simulate the number of groups of a data set, To combine the covariance of the gaussian process model, A matrix of known regression term vectors that combine the gaussian process model, Super parameters to be estimated for the joint gaussian process model; inputting the optimal calibration parameters serving as calibrated calibration parameters and model variable parameters to be predicted into a combined Gaussian process model to obtain a calibrated response predicted value, wherein the response predicted value is expressed as: , Wherein, the As a result of the calibrated standard parameters, 、、 Respectively estimated 、、 , In order to calibrate the response prediction value, For model variable parameters Is provided.
2. The machine learning-based unmanned wheel ground coupling model optimization method of claim 1, wherein the steps of acquiring all-terrain unmanned wheel ground coupling related data, dividing the all-terrain unmanned wheel ground coupling related data into model variable parameters and calibration parameters, and constructing an all-terrain unmanned wheel ground coupling simulation model specifically comprise: Based on all-terrain unmanned wheel ground coupling related data, dividing the model variable parameters into model variable parameters which can be observed experimentally and calibration parameters which cannot be observed experimentally, wherein the calibration parameters comprise at least one of soil shear modulus, soil cohesion, wheel-soil contact friction coefficient, soil inter-particle recovery coefficient and soil inter-particle static friction coefficient, and the model variable parameters comprise at least one of wheel structure size, vehicle body quality, suspension system parameters and running speed; setting a value interval of the calibration parameters according to expert information or engineering experience information; And constructing a wheel-ground contact simulation model by a discrete element method, and combining an unmanned vehicle dynamics model constructed by multi-body dynamics simulation software to construct an all-terrain unmanned wheel-ground coupling simulation model.
3. The machine learning based unmanned wheel ground coupling model optimization method of claim 1, wherein the step of constructing a gaussian process proxy model of a simulation model based on the simulation dataset, estimating super parameters of the gaussian process proxy model by using a maximum likelihood estimation method, and outputting simulation response predicted values of non-sampling points specifically comprises: based on the simulation data set, a Gaussian process proxy model of a simulation model is constructed, which is expressed as follows: , , , , , , Wherein, the In the case of a gaussian process model, To simulate the output response values of a data set, The model variable parameters entered for the simulation dataset, The calibration parameters entered for the simulation data set, As a mean function of the gaussian process proxy model, A matrix of known regression term vectors for the gaussian process proxy model, For the hyper-parameters to be estimated in the gaussian process proxy model, For simulating data sets 、 The covariance of the two sets of inputs, For the variance to be estimated, As a correlation matrix, a correlation matrix is used, Is the first The correlation of the individual variables is to be estimated by the hyper-parameters, For the super-parameter set for which the correlation is to be estimated, 、 Respectively the first A group(s), A set of simulation data sets is provided, Is that 、 The degree of correlation between the two, 、 Respectively the first A group(s), Group simulation dataset first The number of variables that can be used, Is the total number of variables; Estimating super parameters of the Gaussian process agent model by adopting a maximum likelihood estimation method, wherein the super parameters are expressed as follows: , , Wherein, the To be estimated, the generic term of the hyper-parameters to be estimated in the Gaussian process proxy model of the simulation model, namely 、、 , Is that Likelihood functions of (a) are provided.
4. The machine learning based unmanned wheel ground coupling model optimization method of claim 3, wherein the step of constructing a gaussian process model of a modeling bias function based on modeling bias between the experimental dataset and the simulated response predicted value, and establishing an association with the gaussian process proxy model and experimental error parameters specifically comprises: Based on modeling deviation between the experimental data set and the simulation response predicted value, constructing a Gaussian process model of a modeling deviation function, which is expressed as follows: , Wherein, the To model the gaussian process model of the deviation function, For model variable parameters in experimental data set Is used for the modeling of the deviation function of (c), For the mean function of the gaussian process model modeling the deviation function, Covariance of the gaussian process model modeling the bias function; And establishing an association relationship by combining the Gaussian process proxy model and experimental error parameters, wherein the association relationship is expressed as: , Wherein, the From the slave Or (b) The value of the product is taken out, For the estimated correlation coefficient(s), The response value is output for the predicted simulation, For the predicted modeled deviation value(s), Is the experimental error.
5. The machine learning based unmanned wheel ground coupling model optimization method of claim 4, wherein the step of integrating the simulation dataset with the experimental dataset to construct a joint gaussian process model, specifically comprises: integrating the simulation data set and the experimental data set to obtain a combined data set, which is expressed as: , Wherein, the In order to combine the data sets, Input parameter space and output response for the joint dataset; a gaussian process model based on the gaussian process proxy model and the modeled deviation function is expressed as: , , , , Wherein, the As a mean function of the joint gaussian process model, 、 A matrix of known regression term vectors of the gaussian process proxy model of the simulation model, the gaussian process model of the modeling deviation function respectively, For proxy model of gaussian process 、 The covariance of the two sets of inputs, 、 Respectively experimental data sets 、 Two sets of input combinations of calibration parameters to be calibrated As an input to the gaussian process proxy model, Is that × Is used for the matrix of units of (a), As the standard deviation of the experimental error, Is the covariance of the joint gaussian process model.
6. A machine learning based unmanned wheel ground coupling model optimization system for implementing the machine learning based unmanned wheel ground coupling model optimization method of any one of claims 1 to 5, the system comprising: The data acquisition module is used for acquiring all-terrain unmanned wheel ground coupling related data, dividing the all-terrain unmanned wheel ground coupling related data into model variable parameters and calibration parameters, and constructing an all-terrain unmanned wheel ground coupling simulation model; The data processing module is used for sampling in a value interval of the model variable parameter and the calibration parameter by using a Latin hypercube sampling method, driving the simulation model to simulate in batches to obtain a simulation data set, sampling in the value interval of the model variable parameter, carrying out a real vehicle test by combining a preset driving working condition, and collecting observation data to form an experimental data set; The Gaussian process agent model construction module is used for constructing a Gaussian process agent model of a simulation model based on the simulation data set, estimating super parameters of the Gaussian process agent model by adopting a maximum likelihood estimation method, and outputting simulation response predicted values of non-sampling points; the association relation construction module is used for constructing a Gaussian process model of a modeling deviation function based on modeling deviation between the experimental data set and the simulation response predicted value, and establishing association relation by combining a Gaussian process proxy model and experimental error parameters; The combined Gaussian process model construction module is used for integrating the simulation data set and the experimental data set to construct a combined Gaussian process model; The model optimization module is used for constructing a joint likelihood function based on the joint Gaussian process model, maximizing the joint likelihood function through an optimization algorithm, solving optimal calibration parameters, super parameters and correlation coefficients of the Gaussian process model modeling a deviation function, and comprises the following steps: Constructing a joint likelihood function based on the joint Gaussian process model, and maximizing the joint likelihood function through an optimization algorithm, wherein the joint likelihood function is expressed as: , , Wherein, the In order to combine the likelihood functions, For the number of sets of experimental data sets, To model the hyper-parameters to be estimated of the gaussian process model of the deviation function, To summarize the estimated hyper-parameters in the gaussian process proxy model, In order to combine the data sets, In order to combine the output responses of the data sets, As the calibration parameter to be calibrated, For the correlation coefficient to be estimated, In order to simulate the number of groups of a data set, To combine the covariance of the gaussian process model, A matrix of known regression term vectors that combine the gaussian process model, Super parameters to be estimated for the joint gaussian process model; the prediction output module is used for inputting the optimal calibration parameters serving as calibrated calibration parameters and model variable parameters to be predicted into a combined Gaussian process model to obtain a response predicted value after calibration, and the response predicted value is expressed as: , Wherein, the As a result of the calibrated standard parameters, 、、 Respectively estimated 、、 , In order to calibrate the response prediction value, For model variable parameters Is provided.
7. A readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the machine learning based unmanned wheel-ground coupling model optimization method according to any one of claims 1 to 5.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the machine learning based unmanned wheel-ground coupling model optimization method of any of claims 1 to 5 when the program is executed.

Description

Unmanned wheel ground coupling model optimization method and system based on machine learning Technical Field The invention relates to the technical field of simulation, in particular to an unmanned wheel ground coupling model optimization method and system based on machine learning. Background The application of all-terrain unmanned vehicles in the fields of military reconnaissance, emergency rescue, geological exploration, unmanned agriculture and the like is continuously deepened, and the high-precision simulation of the mechanical behavior of the all-terrain unmanned vehicles interacting with the ground of an off-road environment is increasingly important. The wheel coupling simulation model breaks through the traditional rigid ground assumption, can capture complex mechanical behaviors such as elastic deformation of tires, compaction and shearing of soil and the like, but the existing model has two major core problems, namely modeling deviation caused by assumption rationality, structural fineness, boundary condition simplification and the like in the modeling process, limited parameter measurement means and dynamic evolution of a terrain medium, so that core parameters are difficult to be matched with a real scene, and finally the deviation between simulation output and actual mechanical behaviors is larger. The traditional calibration method is highly dependent on priori distribution information of calibration parameters, the parameters are forcedly brought into standard models such as normal distribution, uniform distribution and the like, the parameters are disjointed with complex and changeable actual operation environments, the calibration result cannot reflect the actual wheel-ground coupling mechanical behavior, meanwhile, the traditional method cannot distinguish inherent deviation of a quantized model and experimental observation errors due to the fact that the traditional method comprises simplifying assumptions, the inherent deviation and experimental observation errors are all attributed to parameter deviation, the calibration result deviates from physical characteristics due to the fact that the calibration parameters are excessively optimized, and the generalization capability of the model is reduced. Therefore, a calibration method and system are needed to solve the problem of insufficient prediction accuracy of the existing model. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide an unmanned wheel ground coupling model optimization method and system based on machine learning, which aim to solve at least one problem in the background art. A first aspect of the present invention is to provide a machine learning-based unmanned wheel ground coupling model optimization method, the method comprising: Acquiring all-terrain unmanned wheel ground coupling related data, dividing the data into model variable parameters and calibration parameters, and constructing an all-terrain unmanned wheel ground coupling simulation model; Sampling in a value interval of model variable parameters and calibration parameters by using a Latin hypercube sampling method, driving a simulation model to perform batch simulation to obtain a simulation data set, sampling in the value interval of the model variable parameters, carrying out a real vehicle test in combination with a preset driving working condition, and collecting observation data to form an experimental data set; Based on the simulation data set, constructing a Gaussian process proxy model of a simulation model, estimating super parameters of the Gaussian process proxy model by adopting a maximum likelihood estimation method, and outputting a simulation response predicted value of an un-sampled point; Based on modeling deviation between the experimental data set and the simulation response predicted value, constructing a Gaussian process model of a modeling deviation function, and establishing an association relationship by combining a Gaussian process proxy model and experimental error parameters; Integrating the simulation data set and the experimental data set to construct a combined Gaussian process model; Constructing a joint likelihood function based on the joint Gaussian process model, maximizing the joint likelihood function through an optimization algorithm, and solving optimal calibration parameters, super parameters of the Gaussian process model modeling a deviation function and correlation coefficients; and inputting the optimal calibration parameters serving as calibrated calibration parameters and model variable parameters to be predicted into a combined Gaussian process model to obtain a response predicted value after calibration. According to one aspect of the above technical solution, the steps of obtaining all-terrain unmanned wheel ground coupling related data, dividing the data into model variable parameters and calibration parameters, and constructing an all-terrain unmanned wheel ground cou