Machine Learning Model for a Biocontact Oxidation Process Driven by Battery-Free Wind-Solar Power Generation─A New Path for Rural Sewage Treatment

Abstract

The energy and operation costs have always been a bottleneck, restricting the development of rural sewage treatment. This work proposes a biocontact oxidation process driven by battery-free wind-solar power generation to implement the automated operation of rural sewage treatment. An automatic machine learning model was designed to predict the performance of the system. Higher accuracy was obtained using the random standard deviation sampling (RSDS) data amplification method. Correlation analysis indicated that influent quality indicators had a path coefficient of 0.899 (p < 0.001) and a significant contribution of 38% for effluent quality variation. Spearman correlation between each influent and effluent quality indicator was greater than 0.9 (p < 0.001), combining variable importance analysis of the model, showing that influent quality indicators presented higher importance than meteorological conditions and power generation capacity. The dependence relationships of important variables on the predicted values and the model stability were further examined using interpretability partial dependence (PD) and individual conditional expectation (ICE) analyses. This study proposed an environmentally friendly rural domestic sewage treatment system with a data-driven system operation model, which provides novel ideas for improving low-carbon and intelligent rural environmental sanitation in the future.