Accurate Prediction of Plant-Scale Biogas Production Based on Multiple Hybrid Machine Learning

Abstract

Plant-scale food waste anaerobic digestion parameters are highly nonlinear and suffer from data imbalance problems, while traditional machine learning algorithms are unable to accurately learn from imbalanced data, resulting in low prediction accuracy. Therefore, this paper proposes an extreme learning machine (ELM) model based on the regression-based synthetic minority oversampling technique (SMOTER) and a genetic algorithm (GA), which is expected to improve the prediction accuracy of the model by solving the AD data imbalance problem. The results showed that SMOTER-GA-ELM had the best prediction accuracy, with a prediction of R 2  0.972 for validation data. In addition, feature importance analysis showed that feed loading and Volatile fatty acids of anaerobic digestion were the two most important factors. Finally, the model was developed as a software and measured using independently collected plant data with an average error of 2.1557%, allowing for accurate prediction and production guidance for plant biogas production.

Keywords: anaerobic digestion, Machine Learning, Predict biogas production, Graphical User Interface Software