Apply machine learning to predict greenhouse gas emissions in aerobic composting and achieve emission reduction by nanomembrane covering mode

Previous studies have utilized nanomembranes to reduce greenhouse gas (GHG) emissions in composting. However, few studies have explored their effectiveness in cold environments for reducing both GHG and NH₃ emissions. This study applied machine learning model to predict GHG and NH₃ emissions in composting and to evaluate nanomembrane technology in aerobic composting under cold conditions. We tested four models: multilayer perceptron (MLP), support vector regression (SVR), random forest (RF), and k-nearest neighbors (KNN). The MLP model achieved the best performance for NH₃ prediction, with an R² of 0.8954, an RMSE of 6.3531, and an MAE of 2.7755. It predicted GHG emissions for Heilongjiang Province. The SHapley Additive exPlanations analysis identified total nitrogen (TN), temperature, and moisture content (MC) as the most influential factors affecting gas emissions during composting. By inputting the experimentally derived feature values into the MLP model, we compared the predicted values of GHG and NH₃ with actual measurements under nanomembrane-covered conditions to quantify emission reduction. This study offers technical methods for manure fertilization and emission reduction in cold environments and provides a valuable reference for the application of aerobic composting mode with nanomembrane in the cold high-altitude regions of Northeast China.