Novel Intelligent System Based on Automated Machine Learning for Multiobjective Prediction and Early Warning Guidance of Biogas Performance in Industrial-Scale Garage Dry Fermentation

Industrial-scale garage dry fermentation systems areextremelynonlinear, and traditional machine learning algorithms have low predictionaccuracy. Therefore, this study presents a novel intelligent systemthat employs two automated machine learning (AutoML) algorithms (AutoGluonand H2O) for biogas performance prediction and Shapleyadditive explanation (SHAP) for interpretable analysis, along withmultiobjective particle swarm optimization (MOPSO) for early warningguidance of industrial-scale garage dry fermentation. The stackedensemble models generated by AutoGluon have the highest predictionaccuracy for digester and percolate tank biogas performances. Basedon the interpretable analysis, the optimal parameter combinationsfor the digester and percolate tank were determined in order to maximizebiogas production and CH4 content. The optimal conditionsfor the digester involve maintaining a temperature range of 35-38 degrees C, implementing a daily spray time of approximately 10 min anda pressure of 1000 Pa, and utilizing a feedstock with high total solidscontent. Additionally, the percolate tank should be maintained ata temperature range of 35-38 degrees C, with a liquid level of1500 mm, a pH range of 8.0-8.1, and a total inorganic carbonconcentration greater than 13.8 g/L. The software developed basedon the intelligent system was successfully validated in productionfor prediction and early warning, and MOPSO-recommended guidance wasprovided. In conclusion, the novel intelligent system described inthis study could accurately predict biogas performance in industrial-scalegarage dry fermentation and guide operating condition optimization,paving the way for the next generation of intelligent industrial systems.