Research On Semi-Active Suspension Control Method Based On Data-Driven Homotopic Reinforcement Learning

With the development of economy and technology,people have higher requirements for the stability of the ride.vehicle suspension system plays a vital role in the stability and safety of the ride.Passive suspension that cannot adjust spring stiffness and damping cannot meet people’s requirements for vehicle dynamic performance while the emergence of the active suspension has changed these drawbacks.Active suspension relies on external energy to adjust the control power acting on the car body.It can change the damping and stiffness of the suspension system according to different road conditions,so as to slow down the body vibration and improve the stability of driving.As a kind of active suspension,semi-active suspension is widely used and attracts more and more attention due to its cost advantage.Active suspension produces different effects under different control algorithms and how to design more advanced and more intelligent control algorithms has always been the focus of relevant scholars.The study of the control algorithm of the semi-active suspension system has important theoretical value and economic benefit,which is also the subject of this paper.For the continuous-time nonlinear semi-active suspension system,this paper first establishes a quarter automobile suspension model of two degrees of freedom,and determines the performance evaluation index of the control system.According to the non-linear characteristics of the semi-active suspension system,the T-S fuzzy model of the suspension system is established.The original nonlinear system model is decomposed into multiple linear subsystems by fuzzy algorithm.Then for the optimal control problem of semi-active suspension,a model-free homotopic reinforcement learning algorithm based on data-driven is proposed with no given initial stability control strategy is required.The state and control input data of each subsystem are measured under the specified fuzzy rule,and the optimal control gain of each subsystem is calculated using the measurement data.We obtain the optimal control strategy by the parallel distribution compensation method.Finally,a simulation analysis of the proposed control strategy is performed using the Matlab software.It is shown that the proposed control strategy can move the unstable poles of the closed-loop system to stable regions with unknown system model information by analyzing and processing the sampled data.Both the system output and the control inputs converge rapidly to zero given the initial value.Comparing the control gain and the optimal control gain,we prove that the control strategy derived by the algorithm is the optimal control strategy.The ideal control effect is achieved in controlling the model-free continuous-time nonlinear system,which can effectively improve the ride comfort of the car.