Research and Application on Internal Model Control Algorithm Based on Echo State Network

摘要

Internal model controller(IMC) is widely used due to its simple structure and flexible parameter adjustment. However, the inverse model of the object is difficult to obtain accurately, especially for some non-minimum phase systems. To solve this problem, this paper proposed an neural network IMC combining Echo State Network(ESN) and IMC method. The initial ESN inverse controller and reference model are trained offline based on input and output data. In order to detect the model mismatch, the mutual information between input and model deviation are calculated. The online update of controller and reference model are achieved based on Recursive Least Square method to enhance better robustness. Simulations and experiments shows the effectiveness of this method and it has some ability of self diagnosis and self recovery.