Daily gas production prediction for a single CBM well using an integrated framework of temporal-multivariate dependency simultaneous modeling: A case study

Accurate prediction of coalbed methane (CBM) production is essential for optimizing production systems and improving development efficiency. However, there are many existing challenges, including complex and low-quality preprocessing, limited inter-variable dependency learning, the requirement for explicit graph construction, and a direct processing incapability for time series data, leading to insufficient data mining and constrained prediction accuracy. To address this issue, this study proposes a deep learning-based integrated framework (MST-DeepGNN) for single-step daily gas production prediction across three types of CBM datasets from the research block. Initially, an adaptive data preprocessing module is developed to reduce preprocessing workload and improve prediction accuracy. It effectively relieves the excessive deviation in imputed data, preserves original statistical attributes, and mitigates the detrimental effects of outliers and non-stationary information on model performance. Subsequently, FourierGNN is introduced to naturally and synchronously capture temporal and inter-variable dependencies, fully exploiting informative patterns in CBM multivariate time series data. Unlike conventional GNNs, it directly processes time series without explicit graph construction or additional temporal modules, offering a unified, simple, and flexible modeling framework. Results demonstrate that MST-DeepGNN realizes higher daily gas production prediction accuracy, more robust generalization capabilities, and more stable performance across three CBM datasets. Compared to the second-best algorithm, MST-DeepGNN achieves average reductions of more than 60 % in MAPE, MAE, and RMSE across three datasets. This study presents an effective deep learning framework for predicting daily gas production from various CBM wells, thereby contributing to the intelligent and efficient development of CBM production systems.