Application of Deep Machine Learning Techniques in Oil Production Forecasting

Abstract

Forecasting of oil production can dearly affect oil field projects because of its economic value. It also decides project planning for the future. Established calculations including nodal analysis, numerical simulation, and decline curve analysis (DCA) require normalized data types. Curve fitting process is used by reservoir and production engineer manually for such analysis and prediction to get the outcome. Availability of limited data, heterogeneity and the complexity of the reservoir can lead to erroneous prediction using conventional methods. This process can be tedious and time consuming while evaluating large amount of data. Solution and the best results are getting from machine learning techniques. In this paper, neural network by computation (CNN) and recurrent neural network (RNN) and long short-term memory (LSTM) model is based on various sequencing structure of input and output steps. In this study, 5 production oil wells data of Indian fields have been used to perform forecasting using above machine learning techniques of time series analysis. The output from the machine learning methods compared with decline curve method. The root mean square error is found to be as low as 0.918 which shows accuracy of deep ML methods. It is also found that CNN model is faster than RNN and DCA in oil production forecasting.