A comprehensive machine learning-coupled response surface methodology approach for predictive modeling and optimization of biogas potential in anaerobic Co-digestion of organic waste

The increasing demand for renewable energy sources has driven the research and development of biogas production as a sustainable and efficient solution. Biogas production depends on various process parameters, which are complicated and highly non-linear processes that need to be controlled at optimal levels to ensure high productivity. The current study seeks to predict and optimize the biogas production yield in an anaerobic digester using co-digestion. Using the central composite design (CCD), L30 orthogonal arrays were developed for experimentation at four factor five levels of parameters: solid concentrations (5-25 %), pH levels (4-8), temperatures (30-50 degrees C), and co-digestions (0-40 %). Furthermore, three different gradient boosting algorithms (AdaBoost, Light Gradient Boosting Machine, and Extreme Gradient Boosting) were used to develop an ML-based prognostic model using experimental data. The developed prognostic model was used to predict the biogas yield. Three performance indicators (R2, RMSE, and MAE) were used to evaluate the robustness of the algorithms. Based on the coefficient of determination (R2 = 0.999), RMSE (0.6265), and MAE (0.4669), the XGB model has the most accurate predictions, followed by LGBM (R2 = 0.996, RMSE = 1.488, MAE = 1.1316), and then AdaBoost (R2 = 0.988, RMSE = 8.8943, MAE = 7.265). To optimize the parameters of biogas production, the Response Surface Methodology (RSM)-based desirability approach was employed. The finding showed that all of the considered parameters significantly affected the biogas yield. The optimal combination of the findings included a solid concentration of 11.44 %, a pH of 6.96, a temperature of 38.94 degrees C, and a co-digestion rate of 39.0 %, which led to a biogas yield of 6029.28 ml. The RSM finding was confirmed by an experimental investigation, and the error rate was within an acceptable limit. It was observed that RSM coupled with machine learning is an effective hybrid technique for modeling, predicting, and optimizing biogas production yield.

Automated summary

The study focuses on predicting and optimizing the yield of biogas production in an anaerobic digester using co-digestion. Experimental data was used to develop a machine learning-based prognostic model, and the Response Surface Methodology (RSM) was employed to optimize the parameters. The results demonstrate that RSM coupled with machine learning is an effective technique for modeling, predicting, and optimizing biogas production yield.