An efficient self-organizing RBF neural network for water quality prediction

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Science

Background:

• Accurate water quality prediction is crucial for environmental monitoring and wastewater treatment.
• Traditional neural networks often lack adaptability, leading to suboptimal performance and efficiency.
• Dynamic adjustment of network complexity is needed for robust prediction models.

Purpose of the Study:

• To introduce a novel Flexible Structure Radial Basis Function Neural Network (FS-RBFNN).
• To enable dynamic structural adaptation for maintaining prediction accuracy and computational efficiency.
• To apply the FS-RBFNN for effective water quality prediction in wastewater treatment.

Main Methods:

• Development of a Flexible Structure Radial Basis Function Neural Network (FS-RBFNN).
• Online addition/removal of hidden neurons based on neuron activity and mutual information (MI).
• Analysis of algorithm convergence during dynamic and post-modification phases.
• Testing on nonlinear dynamic system identification and wastewater treatment water quality prediction.

Main Results:

• The FS-RBFNN achieved efficient RBF structure design with fewer hidden neurons.
• Significantly reduced training times compared to other algorithms.
• Demonstrated high effectiveness in predicting water quality in wastewater treatment processes.
• Maintained prediction accuracy through dynamic structural adjustments.

Conclusions:

• The proposed FS-RBFNN offers an effective approach for adaptive neural network design.
• Dynamic structural modification enhances both prediction accuracy and computational efficiency.
• The FS-RBFNN is a promising tool for real-world water quality prediction applications.