Research On Driving Parameter Estimation Based On Neural Network Of Four-in-Wheel Drive Electric Vehicle

Four-wheel drive electric vehicles have become an important research direction in the field of electric vehicles due to the advantages of energy conservation,environmental protection and vehicle dynamics control.Furthermore,on the basis that we can access the vehicle state parameters real-time and accurately,then the control objective of vehicle control systems can be achieve.Due to the limits of cost and practical application conditions,control system cannot obtain sufficient vehicle state parameters,which reduces the control performance of the vehicle.Therefore,this paper takes the four-wheel drive electric vehicle as the research object,and studies in two directions of " vehicle sideslip angle estimation" and "road adhesion coefficient estimation".The specific research contents are as follows:Firstly,the vehicle dynamics model and simulation model are established.The seven degree of freedom vehicle model which can accurately reflect the movement of vehicle body and wheels is established;Vehicle dynamics model can accurately reflection the motion of vehicle body、suspension and wheel are established based on Maple Sim.The following hardware-in-the-loop simulation platform(Hi L)is built and the Hi L verification of the estimation algorithm basis on Maple Sim multi-body vehicle dynamics model.Secondly,side slip angle is a important parameter which represents the state of vehicle.In this paper a novel combined corrective kinematic method which based on the steady-state model of the side slip angle and the RBF neural network is proposed.In this article a steadystate model of the slip angle is constructed based on the 2-Do F vehicle lateral model.To obtain tire cornering stiffness through the steady-state model,we design a tire cornering stiffness observer to estimate based on the ridge regression method;Then,the proposed model is utilized to corrective kinematic method for estimation of the lateral velocity.The RBF neural network is used to correct the kinematics method to achieve the estimation of the longitudinal velocity.Kinematics expression of decoupling side slip angle of Longitudinal Velocity.Finally,verify the proposed method under normal and combined conditions.The estimation results are statistically analyzed,and the mean,variance and root mean square error of the analysis results prove the effectiveness of the proposed method.Thirdly,the road adhesion coefficient is a important parameter that affects the tire force and the vehicle state estimator above is designed based of it is known.However,it is difficult to obtain the accurate value of road adhesion coefficient in actual vehicles.In this paper a novel combined method which based on the deep neural network and extended state observer is proposed.The method of deep neural network image recognition is used to identify the road state,the PSPNet semantic segmentation model is used to segment the road image,and the road type(dry,wet,snow)of the segmentation result is identified that obtain the road type and driving area;Due to the image recognition method cannot provide accurate road adhesion coefficient value,the road adhesion coefficient estimation method based on the extended state observer is designed according to the wheel model;Combining the characteristics of the two estimation methods,a fusion estimation method with steering angle as the switching condition is designed,the test results prove the effectiveness of the proposed road adhesion coefficient method.Finally,in order to further verify the accuracy and real-time performance of the estimation method,this paper uses NIPXI,d SPACE,driving simulator and other real-time hardware systems to build a hardware-in-the-loop simulation platform,the platform is used to prove the effectiveness of the proposed the vehicle state parameter estimation algorithm.The estimation results are statistically analyzed,and the mean,variance and root mean square error.The analysis results show that the proposed estimation method effectively solve the steady state drift problem of the kinematic estimation method,and solve the problem that the steady-state model estimation method is insufficient in expressing transient performance.The vehicle sideslip angle and vehicle speed are effectively estimated on the road of high and low adhesion coefficients.And the analysis results show that the estimated results of the road adhesion coefficient meet the requirements of the above estimator.