Abstract
This paper presents a novel two-layer intelligent learning control framework using an output recurrent fuzzy neural LSTM broad learning system (ORFNLSTM-BLS) and a RMSprop optimization algorithm, designed to enhance learning capability and address the limitations of existing fuzzy BLS approaches. The first layer focuses on real-time system identification, leveraging ORFNLSTM-BLS to learn complex nonlinear system dynamics and effectively capture long-term dependencies in sequential data. The second layer implements an intelligent adaptive controller based on the ORFNLSTM-BLS, which dynamically adjusts control actions for time-varying setpoint tracking and disturbance rejection. The RMSprop optimization algorithm is applied across both layers to adaptively regulate learning rates, ensuring faster convergence and mitigating gradient issues. The convergence of the proposed method is provided through five theorems, ensuring gradient boundedness and learning rate conditions. Additionally, a comprehensive study on time complexity and memory consumption is presented to guide controller board selection. Simulations and experimental results demonstrate the superiority of the proposed framework in achieving accurate control performance, robustness under uncertainty, and real-world applicability, where conventional PID control fails to meet the required system response. This method offers valuable insights for developing advanced intelligent control solutions in dynamic environments involving sequential data and uncertainty.
Funding Information
  • National Science and Technology Council (NSTC), Taiwan (NSTC 112-2221-E-005-077)

This publication has 32 references indexed in Scilit: