Optimizing swine manure composting parameters with integrated CatBoost and XGBoost models: nitrogen loss mitigation and mechanism

In this study, machine learning was used to optimize the aerobic composting process of swine manure to enhance nitrogen retention and compost maturity in order to meet the demand for high-quality organic fertilizers in sustainable agriculture. In this paper, multidimensional parameter data of swine manure composting were collected, six machine learning models (including CatBoost and XGBoost) were constructed, and the model parameters were optimized by genetic algorithm. Through model interpretation analysis (SHapley Additive exPlanations and Partial Dependency Plots), experimental validation and mechanism study, the significant effects of operating parameters on composting process and nitrogen loss were revealed. The results showed that optimal control of moisture content, compost temperature and aeration could effectively improve compost quality (GI nearly 198 %), reduce NH₃-N and N₂O-N emissions by 35.17 % and 9.70 %, and promote nitrogen conversion by increasing microbial community activity. This approach provides a new way for the efficient resource utilization of agricultural waste, which can help reduce the dependence on chemical fertilizers.