Adaptive linearization control based on reinforcement learning

Based on the feedback linearization theory, this paper demonstrates how a reinforcement learning scheme, adopted to construct an artificial neural network (ANNs), can linearize a nonlinear system more effectively and cancel the effect of nonlinearity of a plant. The proposed reinforcement linearization learning system (RLLS) consists of two sub-systems: one is a long-term policy selector, evaluation predictor (EP) element, and the other is a short-term action selector, consisting of linearizing control (LC) and reinforce predictor (RP) elements. In addition, an affine linear reference model plays a role of the environment (instructor), which provides the reinforcement signal in the linearizing process. Simulation results demonstrate that the proposed scheme has the better performance of reliability and robustness of the controlled structure than conventional ANNs schemes.