Development of a Machine Learning Technique in Conjunction with Reservoir Complexity Index to Predict Recovery Factor Using Data from 18,000 Reservoirs

Abstract

With increased popularity and success of Machine-Learning (ML) Techniques, there is continued interest in developing new tools while exploring the boundaries of the technique. Here we use ML techniques to identify reservoir engineering, geological and development features that influence the ultimate recovery factor in a group of reservoirs, and to develop a model that provides the relation between recovery factor (RF) and these influencing features. Furthermore, we use techniques that allow opening the "black-box model" to decipher the relation between the model output and the individual features. The overall process included: (i) Creating a subset of reservoirs from the main database that included high quality reliable data, (ii) using SME (Subject Matter Expert) knowledge to develop a list of expected influential attributes, (iii) investigating the quality and the quantity of the data within each attribute and imputing when necessary, (iv) utilizing data analysis tools to reduce the number of input variables, and (v) examining the success of the model in modelling the test data. A Reservoir Complexity Index (RCI) factor was calculated using the 6 attributes identified and input into the ML model. For a dataset of approximately 18,000 reservoirs with high confidence in the recovery factor, a model was developed with 6 numerical features. The model exhibited an accuracy of ±5% for 50% of the data (between P25 and P75) and ±10% for 80% of the data (between P10 and P90). Contrary to our expectations, inclusion of the categorical features available in the reservoirs’ database did not improve the accuracy of the model. While this work falls within a growing category of similar work, its novelty is in demonstrating the strength of XGBoost in modelling and of SHapley Additive exPlanations (SHAP) in deciphering the relative contributions for each attribute of the model. Furthermore, we provided multiple examples to demonstrate and explore both the value and the limitations, and failure of application, of the ML techniques in this kind of work. This learning is important for a cautious but robust adaptation and acceptance of ML techniques.

Investigation and Optimization of EOR Screening by Implementing Machine Learning Algorithms

“…Sinha et al [24] and Mahdaviara et al [25] used reservoir information as input parameters to establish regression or classification models, and the random forest model provided a basis for decision making. In studying the relationship between oil recovery and geological characteristics, researchers established a random forest prediction model to provide a basis for oil reservoir exploitation [26,27]. Yao et al established and compared several machine learning models in the study of the relationship between surfactant concentration, porosity, and permeability, and the random forest model achieves better accuracy [28].…”