Modeling And Analysis Of Driver-Vehicle System Based-on Model Predictive Control

The research of handling stability has been developed from open loop method to closed loop method. At the beginning, much attention has been paid to building accurate vehicle model. But it’s difficult to evaluate the vehicle handling performance and to design a car if just knowing the response of a vehicle under certain input as an open loop system. So now there are more research on building a driver model and car is only the being controlled unit in the driver-vehicle-road closed loop system which is to investigate the property of the whole system.This paper is based on the model predictive control(MPC) to model the steering behavior of drivers with a CarSim vehicle model in a driver-vehicle closed loop system, and to analyze the steering property of drivers and to provide suggestions to human drivers.Firstly, a non-linear 2 degree-of-freedom vehicle model is used as predictive model to imitate the perception of drives on the car. 2 DOF vehicle model contains lateral speed and yaw rate which can be perceived by drivers and are also closely related to handling stability. By using non-linear tire model whose parameters are identified, a more accurate predictive model is built. Since only linear model can be used in MPC algorithm, linearization is done on-line.Secondly, the optimal input and control goal is determined to model drivers’ preview, prediction and optimization behavior. A multi-point preview model is used so that the driver uses road information within preview distance instead of single point preview. Also the influence of preview distance and preview focus point are investigated. The optimization goal is sophistically minimize the trajectory error, steering angle input and steering angle acceleration. By double lane change tests, it proves that the MPC driver model can follow the road input meanwhile having the same steering angle input as real human driver.Thirdly, the test car is parameterized in CarSim to be used as controlled car. And the CarSim model is validated by steering angle input test with well-equipped sensors and improved sensor installation method. Meanwhile double-lane change tests are done to validate the driver-vehicle closed loop system.