Applied Machine Learning and Deep Learning to Predict Oil and Gas Production

Abstract

This study proposed an approach using Regression algorithms and Ensemble methods to leverage the results. A typical Recurrent Neural Networks include Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) and Simple RNN has been developed to predict oil and gas production. The results have shown that machine learning gained good results in the early stage of the production phase and the recurrent neural networks have improved the prediction in the final stage of the production phase. Compare with conventional decline curve analysis (Arps), LSTM has outperformed due to the ability in time-series prediction. In Volve field studies, LSTM is better than reservoir simulation at well F14 and F15 and close to the accuracy of actual values in all five wells. It demonstrates the ability to use Artificial Intelligence in production forecasting. While reservoir modeling needs a great amount of time and effort to construct a reservoir model with extensive domain knowledge of geology, geophysics, petrophysics, reservoir engineering, and reservoir simulation, LSTM only based the historical data to complete the tasks. So, in this circumstance LSTM is working on cutting-edge methods, and need to be studied carefully in the future in order to give better and accurate results.