Machine learning approach for predicting anaerobic digestion performance and stability in direct interspecies electron transfer-stimulated environments

Abstract

Direct interspecies electron transfer (DIET) stimulation in anaerobic digestion (AD) processes by adding conductive materials has been reported to improve process stability and recovery from process imbalance during long-term continuous operations. In this study, machine learning (ML)-based models using three algorithms, namely artificial neural network, support vector machine, and random forest, were constructed to predict AD efficiency in DIET-stimulated environments. The target output variables were the chemical oxygen demand removal efficiency and methane production rate, which are two major parameters to assess AD efficiency and stability. All constructed ML-based models had high prediction efficiencies for both output variables (correlation coefficient > 0.934), because three operational time-based input parameters were used to reflect the acclimation period of microbial communities after the operating conditions were changed. The results of the random forest model showed that the time-based parameter, which was measured from the time of magnetite addition, was the most important input variables. These results suggest the potential of using ML techniques with varied time-based parameters to predict the stability of AD by stimulating DIET.