Output Feedback Speed Control for a Wankel Rotary Engine via Q-Learning

This paper develops a dynamic output feedback controller based on continuous-time Q-learning for the engine speed regulation problem. The proposed controller is able to learn the optimal control solution online in a finite time using only the measurable outputs. We first present the mean value engine model (MVEM) for a Wankel rotary engine. The regulation of engine speed can be formulated as an optimal control problem that minimises a pre-defined value function by actuating the electronic throttle. By parameterising an action-dependent Q-function, we derive a full-state adaptive optimal feedback controller using the idea of continuous-time Q-learning. The adaptive critic approximates the Q-function as a neural network and directly updates the actor, where the convergence is guaranteed by employing novel finite-time adaptation techniques. Then, we incorporate the extended Kalman filter (EKF) as an optimal reduced-order state observer, which enables the online estimation of the unknown fuel puddle dynamics, to achieve a dynamic output feedback engine speed controller. The simulation results of a benchmark 225CS engine demonstrate that the proposed controller can effectively regulate the engine speed to a set point under certain load disturbances.