Machine learning based prediction and parameters optimization for biohydrogen gas production

Authors:

Abstract
Hydrogen production through the dark fermentation pathway offers an energy recovery alternative via wastewater utilization. The study hybridizes experimental and machine-learning investigations to model and optimize the hydrogen yield through seven major operational variables. Four regression models-one each of Support Vector Machine, Decision Tree, Random Forest, and Neural Network are developed and evaluated through rigorous preprocessing and hyperparameter tuning. The Random Forest turned out to be the best-performing model, offering an R 2 of 0.8006 and an accuracy of 0.9171. A differential evolution algorithm was then employed, ranging over the input parameters, to search for the optimum conditions, with the predicted maximum hydrogen yield set at 0.9361 (L H2/L media/hr). These findings establish the application of Machine Learning (ML) in predicting and establishing the optimum conditions for bio-hydrogen production, thus providing a scalable platform for the environmental recovery of energy from wastewater treatment processes.

This publication has 34 references indexed in Scilit: