Lithofacies Based Fluid Prediction Using Machine Learning

Abstract

Abstract Gas, oil and water zones are critical aspects for geoscientists to evaluate. In each well, a geoscientist must determine where such zones are located based on available well data, such as gamma ray, resistivity, density, neutron porosity, etc. This article addresses how to identify the most important gas, oil, and water zones based on these conventional log curvesusing machine learning (ML) technology which utilize different kinds of neural networks algorithms that is increasingly popular in well log analysis (Leila Aliouane, et al. 2013). Traditionally, geoscientists use a combination of different log curves to identify trends to locate gas, oil, and water zones. This methodology typically uses cut-offs and/or crossovers of different log curves and it is not always reliable. The most commonly used method are the crossover of bulk density and neutron porosity and the cutoffs of resistivity, which is really helpful but not very accurate. ML is categorized as supervised learning or unsupervised learningaccording to whether training sample labels are provided (Witten et al., 2005; Kotsiantis et al., 2007).This paper proposes a new methodology that uses MLtechnology to predict gas, oil, and water zones. It first builds an Unsupervised Machine Learning (UML) model to predict the initial lithofacies. It then builds another Supervised Machine Learning (SML) model using different conventional log curves and tested gas, oil, and water zones as inputs, including the initial predicted lithofacies from each well to train the model. The model is then applied to predict the gas, oil, and water zones for all wells, without the need for additional test information. This automated workflow can achieve a very good result easily and accurately because the machine learning technology can fully utilize the input information to find the relationshipbetween the input data and the training data. By applying the lithofacies restriction to limit the output only within the good reservoirs,a very good final result was successfully achieved. At the end of the workflow a 3D model was also created to predict the distribution of the fluid. The 3D model shows clearly the gas, oil and water distribution horizontally and vertically which will help with field development in the future.