Advanced Steady-state Detection in Biogas Plant Data Using Statistical and Machine Learning Techniques

The application of biogas in the industrial energy transition has strongly emerged as a relevant feedstock for chemicals and energy production, especially in the context of biogas reforming. Effective steady-state detection is crucial for optimizing the production processes, where stable operation directly impacts hydrogen yield and system efficiency. This study tests three steady-state detection (SSD) techniques on two datasets from a demo-scale biogas plant: (i) a statistical hypothesis testing approach, (ii) a trend-based sliding window method, and (iii) the machine learning-based isolation forest (IF) algorithm. A parameter sensitivity analysis was performed to optimize each technique's performance on the biogas plant data. Results indicate that the statistical methods are strongly influenced by the operative parameters, while the IF detects simultaneously outliers and transient values. The integration of an IF algorithm in the detection framework is suggested to enhance reliability in real-time monitoring of biogas reforming processes. This study shows the potential of advanced SSD methods in analysing biogas plant operations to improve hydrogen production and optimal control. Future research will focus on developing a unified detection framework and refining machine-learning models for real-time implementation.