Prediction and optimization of nitrogen losses in co-composting process by using a hybrid cascaded prediction model and genetic algorithm

In this study, the effects of co-composting of food waste and poultry waste on nitrogen losses and maturity were investigated. The different mixture ratios were used and the effectiveness of co-composting was compared with mono-composting of each waste. Also, a linear and nonlinear hybrid tool based on a cascaded forward neural network was used to estimate nitrogen losses of all reactors. The proposed hybrid tool produced predictions with mean absolute percentage error (MAPE) values of approximately 1-2% on all data points containing the training, validation, and test datasets. These results can be considered outstanding, especially when compared to Response Surface Methodology (RSM), which produces predictions with MAPE values of approximately 15% on all data points. The optimal values from the genetic algorithm (GA) were for poultry waste of 17.20%, for a duration of 97.64 days. These findings are invaluable, especially when it is costly and difficult to renew the composting process by creating a new experimental setup.

Automated summary

This study investigated the effects of co-composting of food waste and poultry waste on nitrogen losses and maturity. Different mixture ratios were used and the effectiveness of co-composting was compared with mono-composting of each waste. A linear and nonlinear hybrid tool based on a cascaded forward neural network was used to estimate nitrogen losses. The results showed that co-composting had higher prediction accuracy and could update the composting process without creating a new experimental setup.