Incremental model based online dual heuristic programming for nonlinear adaptive control

Abstract Dual heuristic programming has gained an increasing interest in recent years because it provides an effective process for optimal adaptive control of uncertain nonlinear systems. However, it requires an off-line stage to train a global system model from a representative model, which is often infeasible to obtain in practice. This paper presents a new and efficient approach for online self-learning control based on dual heuristic programming. This method uses a recursive least square method to online identify an incremental model of the system instead of a global system model. The presented incremental model based dual heuristic programming method can adaptively generate a near-optimal controller online without a priori information of the system dynamics or an off-line training stage. To compare the online adaptability of the conventional dual heuristic programming method and the newly proposed method, two numerical experiments are performed: an online reference tracking task and a fault-tolerant control task. The results reveal that the proposed method outperforms the conventional dual heuristic programming method in online learning capacity, efficiency, accuracy, and robustness.