Data-Driven Prediction of Biogas Production in Full-Scale Anaerobic Digestion Systems

Biogas production is a complex, dynamic process influenced by numerous operational and biological factors, making data driven forecasting essential for optimizing operational efficiency and ensuring stable energy outputs in industrial biogas plants. This thesis focuses on the development of forecasting models to predict biogas production which, in real settings, may support plant operators in the control of key operational parameters. The study explores and compares a classical time series model, ARIMA, with the advanced machine learning approach Gradient Boosted Decision Trees (GBDT). By leveraging historical operational data collected from an industrial biogas plant, the models aim to provide accurate medium-term forecasts of biogas yield. Each model is evaluated using standard performance metrics complemented by residual analysis and error progression studies to assess model stability and systematic biases. The results demonstrate that the two models have competing performance but overall, GBDT using exogenous information achieve substantially more stable performance than univariate models, underscoring the importance of nonlinear modeling and full operational data. The outcomes of this work are expected to support more data-driven decision-making, improve process control, and enhance the overall efficiency and reliability of biogas production operations.