An Enhanced Full-Form Model-Free Adaptive Controller for SISO Discrete-Time Nonlinear Systems.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence in Control
• Nonlinear System Dynamics

Background:

• Model-free adaptive control (MFAC) is crucial for systems lacking precise models.
• Existing FFMFAC methods require further optimization for complex nonlinear systems.
• Online parameter adjustment and gradient estimation are key challenges in adaptive control.

Purpose of the Study:

• To propose an enhanced full-form model-free adaptive controller (EFFMFAC) for SISO discrete-time nonlinear systems.
• To integrate Long Short-Term Memory (LSTM) neural networks and Fuzzy Neural Networks (FNNs) for improved adaptive control.
• To achieve online tuning of FFMFAC parameters and estimation of pseudo-gradient values without system models.

Main Methods:

• Utilized LSTM networks for online adjustment of critical FFMFAC parameters.
• Employed FNNs for real-time estimation of controller pseudo-gradient values.
• Developed EFFMFAC using only measured input-output data for online neural network training, avoiding offline training and system identification.

Main Results:

• EFFMFAC demonstrated superior performance over existing methods in simulations.
• Achieved significant Root Mean Square Error (RMSE) reductions of 21.69% and 11.21% compared to FFMFAC.
• Ablation analysis and performance indices validated the effectiveness and rationality of the proposed EFFMFAC.

Conclusions:

• The proposed EFFMFAC effectively enhances control performance for SISO discrete-time nonlinear systems.
• Integration of LSTMs and FNNs enables robust online adaptation and accurate gradient estimation.
• EFFMFAC offers a practical, model-free approach for advanced adaptive control applications.