Output Recurrent Fuzzy Neural LSTM-BLS Controller for Nonlinear Digital Time-Delay Dynamic Systems

In this paper, a novel control architecture is presented by integrating an Output Recurrent Fuzzy Neural Long Short-Term Memory (ORFNLSTM) and a Broad Learning System (BLS) for a class of single-input-single-output (SISO) nonlinear dynamic systems. This new controller, abbreviated as the ORFNLSTM-BLS controller, is especially proposed by combining the techniques of deep learning and broad learning method to establish an adaptive intelligent controller with an online deepest gradient descent learning algorithm to online update its weights. The ORFNLSTM-BLS controller aims to improve the performance of the ORFBLS controller by incorporating the memory of LSTM to handle time-series data more effectively. A sufficient condition of the proposed controller is established to accomplish its uniformly asymptotical stability. The effectiveness and superiority of the ORFNLSTM-BLS controller are well exemplified by carrying out one comparative simulation in comparison with a fixed-gain proportional-integral-derivative (PID) controller, an adaptive predictive PID controller augmented with ORFBLS (ORFBLS-APPID), and an existing ORFBLS controller in terms of two types of control performance indexes: the overall performance indexes and transient state performance indexes. The results show that the proposed ORFNLSTM-BLS controller outperforms the three existing control methods. The developed techniques would provide useful references for professionals working in the fields of process and servomechanism control.