Hydrocarbon production dynamics forecasting using machine learning: A state-of-the-art review

Abstract

Accurate prediction of hydrocarbon production is crucial for the oil and gas industry. However, the strong heterogeneity of underground formation, the inconsistency in oil–gas-water distribution, and the complex flow mechanisms make hydrocarbon production forecasting (HPF) difficult, which leads to a high level of uncertainty in the prediction results. The explosion of machine learning (ML) methodologies that are capable of analyzing big data shed new light on HPF using production data.In this article, an in-depth review is provided regarding HPF using ML methodologies. Firstly, the merits and drawbacks of traditional HPF methods are analyzed and summarized. Then, the applications of ML algorithms in HPF are reviewed in detail, especially concentrating on artificial neural network, support vector machine, and ensemble learning. For each algorithm, the basic theory and its variants are first introduced, and its applications in HPF are comprehensively demonstrated subsequently. Finally, this article presents the challenge and prospects of machine-learning-based HPF. Sophisticated ML proxy models can be constructed and employed to deal with an extended type of input data such that improving the efficacy of data utilization. On the other hand, deep learning models designed to handle time-series data can gain more attention. Modeling approaches for multivariate time-series hydrocarbon production data using deep neural networks with similar functionality to LSTM may lead to more accurate and computationally efficient production forecasting.