Research on an online self-organizing radial basis function neural network.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Radial basis function (RBF) neural networks require effective structure design for optimal performance.
• Automated methods for designing RBF neural network architectures are crucial for complex applications.

Purpose of the Study:

• To introduce a new growing and pruning algorithm for automated RBF neural network structure design, termed self-organizing RBF (SORBF).
• To evaluate the performance of the SORBF method in function approximation, dynamic system identification, and wastewater treatment process modeling.

Main Methods:

• Development of a growing and pruning algorithm based on the receptive field radius of RBF nodes.
• Proposal of parameter adjusting algorithms for the entire RBF neural network.
• Application of the SORBF method for biochemical oxygen demand (BOD) concentration prediction in wastewater treatment.

Main Results:

• The SORBF algorithm demonstrated efficiency in optimizing RBF neural network structures.
• The method achieved superior performance in function approximation and dynamic system identification tasks compared to existing algorithms.
• Successful application of SORBF for capturing BOD concentration in a wastewater treatment system.

Conclusions:

• The proposed SORBF method provides an efficient approach for automated RBF neural network structure optimization.
• SORBF offers improved performance over existing algorithms in various benchmark tasks and a real-world application.
• The algorithm shows promise for applications requiring accurate modeling and prediction, such as environmental monitoring.