Influence of the Prediction Model Complexity on the Performance of Model Predictive Anti-jerk Control for On-board Electric Powertrains

Abstract

Anti-jerk controllers compensate for the torsional oscillations of automotive drivetrains, caused by swift variations of the traction torque. In the literature model predictive control (MPC) technology has been applied to anti-jerk control problems, by using a variety of prediction models. However, an analysis of the influence of the prediction model complexity on anti-jerk control performance is still missing. To cover the gap, this study proposes six anti-jerk MPC formulations, which are based on different prediction models and are fine-tuned through a unified optimization routine. Their performance is assessed over multiple tip-in and tip-out maneuvers by means of an objective indicator. Results show that: i) low number of prediction steps and short discretization time provide the best performance in the considered nominal tip-in test; ii) the consideration of the drivetrain backlash in the prediction model is beneficial in all test cases; iii) the inclusion of tire slip formulations makes the system more robust with respect to vehicle speed variations and enhances the vehicle behavior in tip-out tests; however, it deteriorates performance in the other scenarios; and iv) the inclusion of a simplified tire relaxation formulation does not bring any particular benefit.

Comparison of Anti-Jerk Controllers for Electric Vehicles With On-Board Motors

“…The prediction model adopted for the eNMPC anti-jerk implementation ( Fig. 11) derives from [42], which compares the MPC prediction model formulations from [20], [26], [43] and [44], and shows that the best one is a two-inertia model with backlash. Therefore, this formulation has been adopted here, as described by equations (29)-(31):…”