Electric Vehicle Stability Control Design Based On Learning Predictive Control

Four-wheel drive electric vehicles have the advantages of flexible control and high energy conversion efficiency,as they are able to independently deliver power to each wheel.Therefore,they have attracted extensive attention from both industry and academia.However,adding fourwheel drive to electric vehicles increases the control dimension of the controller,and model errors and environmental uncertainties during vehicle operation pose new challenges to vehicle stability control.Thus,coordinating the front wheel steering angle and four-wheel drive torque,considering environmental uncertainties,is the key to stability control of four-wheel drive electric vehicles.To address these issues,this paper proposes a learning-based predictive control strategy for stability control of four-wheel drive electric vehicles.By compensating for model errors and environmental uncertainties using machine learning algorithms and combining model predictive control with multiple inputs and multiple outputs and explicit constraint handling,we achieve coordinated control of multiple actuators for four-wheel drive electric vehicles in uncertain environments.Compared with traditional predictive controllers,this paper integrates learning mechanisms with vehicle stability control,effectively considering the impact of model errors and environmental uncertainties on vehicle stability,further improving vehicle handling stability,and providing new solutions to the problem of vehicle stability control.The main contributions of this paper are as follows:1.A Burckhardt tire model and a seven-degree-of-freedom vehicle model are established to accurately describe the nonlinear and coupling characteristics of tire mechanics and vehicle dynamics under complex vehicle operating conditions.The seven-degree-of-freedom vehicle model takes into account the lateral,longitudinal,yaw,and dynamic characteristics of the vehicle wheels,based on the analysis of the nonlinear characteristics of tires and the dynamic characteristics of vehicles.Through comparison with a two-degree-of-freedom linear model and a high-precision simulation software CarSim,the established seven-degree-of-freedom vehicle model is verified to have higher modeling accuracy and can accurately describe the nonlinear and coupling characteristics of tires and vehicles.2.A nonlinear predictive control strategy is designed to address the stability control problem of four-wheel drive electric vehicles under extreme operating conditions.By considering the control objectives of vehicle safety and comfort,as well as the constraints on the front wheel steering angle,the driving motor,and other actuators,a control objective function and constraint conditions for vehicle stability control are constructed.Based on model predictive control,an optimization problem description of the control problem is constructed and solved by a nonlinear optimization tool CasADI.To verify the effectiveness of the controller,experiments are conducted under different friction coefficients,and the results show that the designed controller can improve vehicle handling stability under different operating conditions compared with a two-degree-of-freedom linear model-based controller.3.A learning-based predictive control strategy is proposed to address the problem of model errors and environmental uncertainties in the predictive model.The Gaussian process regression method is used to model the model errors and environmental uncertainties,and the datamechanism mixed model is constructed by combining the learned model with the mechanistic model.Based on the control objectives of vehicle stability and the constraint conditions of the actuators,a learning-based vehicle stability controller is designed.To address the insufficient data stimulation problem of the machine learning model,the variance output from the vehicle error model is added to the controller objective function,following the principle of minimum variance control,to improve the confidence of the machine learning vehicle error model output.The Simulink-CarSim co-simulation platform is used to verify that the data-mechanism mixed model can effectively reduce the model mismatch caused by environmental uncertainties,and the designed controller has better control performance and adaptability,further improving the handling stability of four-wheel drive electric vehicles.