Production forecasting in carbonate oil reservoirs through integration of deep learning and geological parameters

Deep learning has emerged as a groundbreaking proxy model for predicting well production, holding great promise for accelerating the advancement and management of the oil and gas industry. However, prior research on well production rate prediction often lacks the incorporation of geological models and comprehensive production data, resulting in limited transfer ability. This research introduces the application of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to forecast the production rate of new wells. By integrating multiple well production data, including factors such as time, choke size, pressure, temperature, water, and demulsifier injection volume, as well as geological parameters like permeability and porosity, the proposed models demonstrate the capability to accurately predict well production rates. Furthermore, the incorporation of one-hot encoding and convolution layers enhances prediction outcomes, leading to a significant 0.2 increase in the R 2 value. The feasibility of utilizing adjacent multi-well data for predicting new well production rates is affirmed, with all models achieving R 2 values exceeding 0.96 when employing encoded simulation data. Particularly, the integration of a CNN layer into the LSTM network showcases a significant improvement of 0.2 in the R 2 value, based on actual oilfield data. These findings suggest that additional performance improvements can be realized through the implementation of data coding techniques and the integration of CNN layers into the model architecture.