Longitudinal model identification of multi-gear vehicles using an LPV approach

This paper aims to provide a data-driven approach for modeling the longitudinal dynamics of a typical ground vehicle with a gasoline engine and automatic transmission. In the identification process, a Linear Parameter Varying (LPV) model is considered, whose inputs are throttle level and road grade and whose parameters vary as a function of throttle level and gear number to capture the nonlinear dynamics. Three parametric structures based on time-series modeling (ARX, ARMAX, BJ) are investigated, whose performances are discussed comparatively. In addition to selecting the best structure, an optimization problem is proposed to acquire the optimal model order of these structures. It is shown, using semi-experimental tests on the CarSim® software package, that the dynamics of different vehicles can best be represented by different model orders. To evaluate the versatility and utility of the proposed LPV model, a PI controller is tuned using the model, and the closed-loop performance of the system is compared against the model. It is shown that the LPV model is very accurate in closed-loop settings.