Machine Learning-Based Prediction and Optimisation of Methane Yield from Cattle Manure Anaerobic Digestion and Thermodynamic Analysis of Temperature–pH Interactions

Abstract

The research involved the use of machine learning approaches to assess the impact of temperature and pH on methane yield during anaerobic digestion of cattle manure, as well as finding the best operating conditions. The literature-based data set consisting of 97 experimental points obtained from peer-reviewed publications was applied for building predictive models. Four different regression models – Linear Regression, Polynomial Regression, Support Vector Regression (SVR) and Random Forest – were trained and tested using methane yield as an output variable. Performance of models was estimated by calculating the coefficient of determination (R²), root mean squared error (RMSE) and mean absolute error (MAE). The Random Forest algorithm showed better predictive performance with test-set R² of 0.4476, RMSE of 53.88 L CH₄/kg volatile solids (VS) and MAE of 44.48 L CH₄/kg VS. Robustness of the developed model was also proven by five-fold cross-validation, resulting in average R² value of 0.386 ± 0.171. Feature importance analysis revealed the contribution of pH equal to 70.4% and temperature – 29.6%, thus emphasizing the greater importance of pH for the studied operational parameters range. Response Surface Analysis determined optimal operating conditions as temperature = 53.7°C and pH = 7.17 leading to maximal methane yield of 272.7 L CH₄/kg VS. Also, the interaction between temperature and pH showed the tendency to shift in optimal pH values depending on temperature increase, which coincided with thermodynamics predictions obtained according to van't Hoff equation. In this work, a new approach involving machine learning and thermodynamics has been used for predicting methane yield and optimising temperature and pH for anaerobic digestion of cattle manure.