ĐIỀU KHIỂN LƯU LƯỢNG DÙNG MẠNG NƠ-RON MỜ HỒI QUY

Abstract

With an online training mechanism, the recurrent fuzzy neural network (RFNN) has been successfully verified by simulations. This study conducts experimental investigations with a supervisory control technique by combining a traditional PI controller and a RFNN-based controller (referred to as RFNNC) on the RT020 liquid flow control unit of the Gunt-Hamburg, Germany. The RFNNC is responsible for fine-tuning the system responses, to overcome the limitation of the fixed parameters of the PI controller. Experimental results on the RT020 device show that during the control process, when the RFNNC's parameter updating algorithm converges, the flow response of the RT020 device has a negligible overshoot. In addition, compared with the default PI controller, the RFNNC has contributed to significantly reduce the system settling time.