Analysis of the Shale Gas Production Prediction Model Based on Machine Learning Using Productivity Factors

Owing to the uncertainties caused by complex flow mechanisms in shale gas reservoirs, production prediction using machine learning has recently been performed. However, efficiently collecting and analyzing production-related factors is difficult. In this study, the productivity factors of different criteria were identified using feature selection, and gas production was predicted using the selected factors in various machine learning models. The error rate improved when productivity factors such as total fluid, measured depth, and lower perforation based on the filter method were applied to the random forest model. In addition, the potential for extending the proposed feature selection process and machine learning algorithms in Marcellus Shale was identified, and a reliable production prediction was expected to be possible by utilizing limited variables even at the sites where production information was difficult to obtain.