Achieving high-quality and safe compost: a study on multi-optimization of a non-linear system using machine learning

Composting represents one of the most effective strategies for animal manure utilization, while its yield efficiency and product safety are constrained by critical bottlenecks including excess nutrient loss and persistence of hazardous residues. There is an urgent need for a method that allows for optimization of multi-objectives and accurate prediction of outcomes integrating these factors. The EcoCompost model was constructed using a multilayer feedforward neural network architecture (R²: 0.87-0.93), specifically engineered for the simultaneous and high-accuracy prediction of nutrient loss, greenhouse gas (GHG) emissions, as well as relative antibiotic degradation and heavy metal abatement. Feature analysis identified the carbon to nitrogen (C/N) ratio as the primary driver of carbon and nitrogen loss, whereas composting duration determined GHG emissions and antibiotic degradation. Partial correlation analysis elucidated the trade-off mechanism allowed for maximizing both antibiotic degradation and nutrient retention, by setting appropriate management of aeration rate, moisture content, and initial C/N ratio. Moreover, the bioavailability of heavy metals decreased with increasing temperature, though this effect was less pronounced than that of composting duration in the initial 20 days; notably, this reduction was significantly augmented by the synergistic action of elevated total phosphorus and alkaline pH. Validation experiments confirm that using the model-derived strategy reduced carbon loss (11.9%), ammonia emissions (37.4%), and bioavailable Cu and Zn (5.3%-18.0%); while maintained high total organic carbon (39.8%) and enhanced humification and stability. Collectively, this study developed a multi-objective optimization model that provides settings for efficiently utilizing animal manure to produce high-quality and safe compost.