A dataset for classifying operational states in dry reforming of biogas processes

Abstract

Dry reforming of biogas (DR) converts methane and carbon dioxide into syngas, offering a sustainable solution for hydrogen production and greenhouse gas reduction. This study uses operational data from DR reactor sensors to predict process states: Activation, Reaction, and Irregularity. Nine reaction-specific datasets were analyzed via 11-fold cross-validation, ensuring test data independence. Machine learning (ML) models — k-nearest neighbors (KNN), Quadratic Discriminant Analysis (QDA), Support Vector Machine (SVM), and Random Forest (RF) — were evaluated, with RF performing best (88.40% accuracy, 89.04% F1-score for Irregularity). ML enables efficient monitoring by capturing complex variable relationships and responding to operational changes. Explainability analysis (SHAP and PDP) identified key variables, including record count, humidity, and pressure. The study provides a robust dataset and methodology for predicting DR states using operational data, supporting future research in fault prediction and process optimization. This approach enhances DR reactor control, advancing reliable and sustainable hydrogen production.