Research On Stability Control Method Of Intelligent Vehicle Based On DDPG

Nowadays,the automobile industry is stepping into the era of intellectualization.Compared with traditional automobiles,intelligent automobiles have active steering control system,which can realize active steering without driver control.When the vehicle is unstable,the driver will get the stable yaw moment by correcting the steering wheel,while the intelligent vehicle has no driver in the ring,but the steering wheel can be corrected by coordinating controller.In order to ensure the driving safety and ride comfort of intelligent vehicle,this paper studies the coordinated stability control method of intelligent vehicle.Firstly,the influence of active steering control and direct yaw moment control(DYC)on vehicle stability is studied.Active steering control is beneficial to improving comfort.DYC ensures vehicle safety under extreme conditions.Coordinating active steering and DYC intervention on vehicles can not only ensure vehicle safety but also improve comfort.This paper takes two-degree-of-freedom vehicle model as a reference model.The ideal yaw angular velocity and sideslip angle of the center of mass are calculated by reference model.The Deep Deterministic Policy Gradient(DDPG)algorithm in DRL is used as the coordinated control algorithm.Five vehicle state parameters are selected as input,and additional yaw moment and steering wheel correction angle are used as output.Coordination controller is designed..Then,the relationship and advantages and disadvantages between Q-Learning,Deep Q-Learning and DDPG algorithm are analyzed.DDPG can handle multi-dimensional input and continuous action output tasks,which meets the requirements of the coordinated control algorithm in this paper.The structure and training framework of DDPG algorithm are studied,and the DDPG algorithm model is built by TensorFlow.As the core of DDPG algorithm,the reward function has great effect on coordinated control.In order to ensure vehicle stability,comfort and moderation of control,the design principle of four-level reward function is put forward in this paper,and the reward function of each level is given.Taking CarSim as the simulation and training platform,the sine with dwell test training environment is built,and the simulation training method of DDPG in CarSim environment is studied.Through the simulation test of DDPG model,the performance indexes of vehicles under different control methods are compared and analyzed,and the control effect of the coordinating controller is verified.Finally,the hardware-in-the-loop experimental platform system is designed.The active steering system and braking system of intelligent vehicle are hardware-based by using NI real-time system as the core.Data acquisition program,steering and braking control program and data exchange program between Python and CarSim are compiled by LabVIEW.The control accuracy of active steering system and braking system is tested and analyzed.The bench test verifies the effectiveness of the stability control method of intelligent vehicle based on DDPG.