An adaptive learning rate for RBFNN using time-domain feedback analysis.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Control Systems Engineering

Background:

• Radial basis function neural networks (RBFNNs) are widely applied in pattern recognition, nonlinear identification, control, and time series prediction.
• Analyzing the learning algorithm of RBFNNs within a feedback structure is crucial for understanding their performance and limitations.

Purpose of the Study:

• To analyze the robustness of the RBFNN learning algorithm under input uncertainties, such as noisy perturbations and modeling mismatch.
• To develop an intelligent adaptation rule for the learning rate of RBFNNs to improve convergence speed and ensure stability.

Main Methods:

• The study analyzes the RBFNN learning algorithm in a feedback configuration.
• Robustness is investigated in the presence of input uncertainties.
• An intelligent adaptation rule for the learning rate is developed, utilizing an estimate of error energy.
• L2 stability is guaranteed using the small gain theorem for upper bounding.

Main Results:

• The developed intelligent adaptation rule for the learning rate of RBFNNs leads to faster convergence.
• The proposed method provides guarantees for L2 stability, even with input uncertainties.
• Simulation results validate the theoretical advancements in RBFNN learning algorithms.

Conclusions:

• The intelligent adaptation rule enhances the performance of RBFNNs in feedback systems.
• The approach offers improved convergence and guaranteed stability in the presence of uncertainties.
• This work contributes to more reliable and efficient applications of RBFNNs in complex systems.