Production Optimization in Oil and Gas Wells: A Gated Recurrent Unit Approach to Bottom Hole Flowing Pressure Prediction

Abstract Production Optimization continues to be a priority in the energy space. Several means have been investigated, particularly machine learning in recent times. One of the aspects machine learning has aided is in the area of future bottomhole flowing pressure prediction (BHFP). It lets asset managers know potential problems and make proactive decisions to facilitate optimal production. This study explores the application of Gated Recurrent Units (GRUs) in oil and gas wells to estimate future BHFP by using optimal features. To properly evaluate the BHFP predictive capability of the GRU model, the model is compared with the use of Long Short-Term Memory (LSTM) networks since LSTM also keeps long-term memory, which aids time series prediction capabilities. Using historical well data, GRU and LSTM models were trained with an optimal number of features, selected by their importance from Recursive Feature Elimination. The results showed the GRU models performed better than LSTM models, averaging a 30% approximate decrease across error metrics. With the GRU forecasts enhanced by optimized features, production managers can make prompt, well-informed decisions. This, in turn, makes it easier to control reservoirs precisely, make proactive modifications to well operations, and cut down on expensive downtime.