Title: An echo state network-based feedforward feedback controller for application in dynamic systems control
Authors: Kazuhiko Takahashi; Naoyuki Kita; Miku Sasaki; Reika Kimura; Masafumi Hashimoto
Addresses: Faculty of Science and Engineering, Doshisha University, Kyoto, Japan ' Faculty of Science and Engineering, Doshisha University, Kyoto, Japan ' Faculty of Science and Engineering, Doshisha University, Kyoto, Japan ' Faculty of Science and Engineering, Doshisha University, Kyoto, Japan ' Faculty of Science and Engineering, Doshisha University, Kyoto, Japan
Abstract: This study explored the application reservoir computing, particularly echo state networks (ESNs), to control dynamic systems. The design method of a servo-level controller was proposed, where the ESN matches the objective plant output with the reference output. The ESN was combined with a feedback controller to obtain the control input of the plant. The ESN-based controller was first trained using a linear-regression approach with fixed datasets gathered from the objective plant. Thereafter, feedback error learning was performed during the control process in real-time to compensate for the control error due to the identification error of the plant's inverse transfer function. Computational experiments involving the control of a discrete-time nonlinear plant were conducted. The simulation results clarified the feasibility of the proposal and validated the performance of the ESN-based controller.
Keywords: reservoir computing; echo state network; ESN; control system; feedforward feedback controller.
DOI: 10.1504/IJAMECHS.2024.137559
International Journal of Advanced Mechatronic Systems, 2024 Vol.11 No.1, pp.40 - 49
Received: 19 Mar 2023
Accepted: 07 Dec 2023
Published online: 25 Mar 2024 *