Multi State Perception And Control Method Of Four Wheel Independently Actuated Electric Vehicles Considering Communication Delays And Sensor Noises

Fully electric vehicles(EV)with individually controlled motors,namely four wheel independently actuated(FWIA)EVs,allow a significant improvement in motion performance due to their remarkable actuation flexibility.The motion control of FWIA EVs has attracted increasing research efforts.In order to achieve this coordinated control,manufacturers widely adopt in-vehicle networks to transmit data between different electic control units.In-vehicle netwoks can considerably reduces wiring cost,weight and increase reliability.Therefore,the overall architecture of FWIA EVs should be treated as a networked control system rather than a centralized control system.However,network communication and controller computation will inevitably induces time delay.Compared with traditional vehicles,there are more information communicated via networks in FWIV EVs,and the network load is higer.The unknown and probably time-varying delays of networks could degrade the control performance of the entire system or even make the system unstable.Furthermore,the noises in sensory signals have been a problem for a long time.In this paper,the multi-state perception of FWIA EVs is studied.Delays and noises in feedback loop can be successfully overcomed.Simulations with a high-fidelity,CarSim,full-vehicle model are carried out to show the effectiveness of this method.The main research work consists the following parts:(1)A brief dynamic model of FWIA EVs is presented.A.Vehicle longitudianal motion,lateral motion,yaw motion,rotations of four driven wheels and the dynamics of brushless DC motors are considered.(2)The specific Electrical/Electronic architecture of FWIA EVs and Controller Area Network communication protocols are analyzed.On the basis of the analysis,load ratio,worst case response time and the overall time delay of the networked control system can be calculated.(3)A perception system of FWIA EVs based on recurrent neural networks(RNN)and unscented Kalman predictor(UKP)is proposed.Instead of solely relying on sensory feedback,which is temporally lagged and tending to be corrupted by considerable amounts of noise,the brain predicts the sensory consequences of our motor commands as well.The ability of prediction can be learned through the experience of interaction with the world.In this paper,state-of-art RNN architecture and training method is used to learn the dynamic model of FWIAs from raw measurement.Then,prior prediction and sensory signals can be fused by UKP to compensate the time-varying time delays and noises.(4)The multi-state perception system is combined with second order sliding mode control method to regulate the direct yaw moment of FWIA EVs and make the yaw rate track desired value.Simulations are conducted in Simulink based on a high-fidelity full-vehicle model constructed in CarSim.The simulation results validate that the chattering problem of sliding mode control,which is caused by time delays and noises,can be effectively alleviated by this method.