Applications of Machine Learning in Sweet-Spots Identification: A Review

Abstract

Summary The identification of sweet spots, areas within a reservoir with the highest production potential, has been revolutionized by the integration of machine learning (ML) algorithms. This review explores the advancements in sweet-spot identification techniques driven by ML, analyzing 122 research papers published in OnePetro, Elsevier, ScienceDirect, SpringerLink, GeoScienceWorld, and MDPI databases within the last 10 years. The review provides a comprehensive analysis of ML applications in sweet-spot identification and highlights best practices in data collection, preprocessing, feature engineering, model selection, training, validation, optimization, and evaluation. The paper categorizes and discusses the different data types used in ML algorithms into six groups, analyzes the combinations of frequently used data types for training and validation, and visualizes the distribution of input parameters and features within each of the six main categories. It also examines the frequency of target variables used in these models. In addition, it discusses various supervised and unsupervised ML algorithms and highlights key studies offering valuable insights for researchers.