Adaptive recurrent cerebellar model articulation controller for unknown dynamic systems with optimal learning-rates

• Published in: 2004 IEEE International Joint Conference on Neural Networks (IEEE Cat. No.04CH37541) Vol. 2; pp. 885 - 890 vol.2
• Main Authors: PENG, Ya-Fu, LIN, Chih-Min, CHIN, Wei-Laing
• Format: Conference Proceeding
• Language: English
• Published: Piscataway NJ IEEE 2004
• Subjects: Adaptive control; Applied sciences; Artificial intelligence; Ceramics; Computer science; control theory; systems; Control systems; Convergence; Exact sciences and technology; Feedback control; Learning systems; Optimal control; Position control; Programmable control; Stability; Dynamic response; Recurrence; Tracking; Precision engineering; Control synthesis; Neural network; Dynamical system; Modeling; Adaptive method; Gradient descent; Associative memory; Descent method; Dynamic stability; Tracking error; Linear motor; Time analysis; Ceramic materials; Gradient method
• ISBN: 0780383591; 9780780383593
• ISSN: 1098-7576; 1558-3902
• DOI: 10.1109/IJCNN.2004.1380047

In this study, an adaptive recurrent cerebellar model articulation controller (ARCMAC) is designed for feedback control system with unknown dynamics. The proposed ARCMAC has superior capability to the conventional cerebellar model articulation controller (CMAC) in efficient learning mechanism, guaranteed system stability and dynamic response. Temporal relations are embedded in ARCMAC by adding feedback connections in the association memory space so that the ARCMAC captures the dynamic response. The dynamic gradient descent method is adopted to adjust ARCMAC parameters on-line. Moreover, the variable optimal learning-rates are derived to achieve most rapid convergence of tracking error. Finally, the effectiveness of the proposed control system is verified by experimental results of linear piezoelectric ceramic motor (LPCM) position control system. Experimental results show that accurate tracking response and superior dynamic performance can be obtained because of the powerful on-line learning capability of the proposed ARCMAC.