Optimizing neural network algorithms for submerged membrane bioreactor: A comparative study of OVAT and RSM...

Area of Science:

• Computational Intelligence
• Chemical Engineering
• Machine Learning

Background:

• Hyperparameter tuning is crucial for maximizing neural network performance.
• Optimizing hyperparameters for Feed-forward Neural Network (FFNN) and Radial Basis Function Neural Network (RBFNN) is essential for accurate predictions.
• Predicting permeate flux in palm oil mill effluent treatment requires robust models.

Purpose of the Study:

• To propose and evaluate a factorial design of experiment and response surface methodology for optimizing neural network hyperparameters.
• To compare the proposed optimization method with the conventional one-variable-at-a-time approach.
• To enhance the prediction accuracy and efficiency of FFNN and RBFNN models for permeate flux.

Main Methods:

• Applied factorial design of experiment and response surface methodology for hyperparameter screening and optimization.
• Utilized Feed-forward Neural Network (FFNN) and Radial Basis Function Neural Network (RBFNN) models.
• Input variables included permeate pump and transmembrane pressure; hyperparameters optimized for FFNN and RBFNN.

Main Results:

• The proposed methodology achieved optimal hyperparameters, resulting in improved model accuracy and reduced generalization error.
• Simulation results demonstrated over 65% improvement in training performance, with decreased processing time and fewer iterations compared to conventional methods.
• The optimized models showed significant enhancements in predicting permeate flux.

Conclusions:

• The factorial design of experiment and response surface methodology provide an effective approach for hyperparameter optimization in neural networks.
• This methodology offers a more efficient and accurate alternative to traditional tuning methods.
• The proposed approach is versatile and can be applied to various neural network applications for parameter optimization.