Predicting Long-Term Borehole Pressure in Gas Wells using a Machine Learning Approach

Accurate, continuous Bottomhole Pressure (BHP) estimation is vital for optimizing production, effective reservoir management, and ensuring gas well integrity. Traditional downhole measurements involve costly, disruptive shut-ins, impractical for long-term surveillance; while PDHGs are expensive and can fail in harsh conditions. This study presents a robust Temporal Convolutional Network (TCN) methodology for 12-month BHP forecasting, focusing on future unseen data from existing wells. TCNs are hypothesized to excel at capturing long-range temporal dependencies due to their parallelism and flexible receptive fields. We benchmark TCNs against XGBoost, RNNs, and LSTMs using historical well data. This research, supporting "Digital Twin" initiatives, aims to develop models for sustained predictive accuracy, offering insights for data-driven BHP prediction, proactive reservoir management, and reduced operational disruptions, highlighting TCNs' potential for continuous well monitoring.