Drilling Data Based Approach for Real-Time Gamma Ray Prediction in Surface Holes Using Artificial Intelligence

Abstract

ABSTRACT: Gamma-ray (GR) logging involves the measurement of natural formations' radioactivity, serving as a valuable tool for distinguishing various lithologies and correlating zones. A prevalent approach is to record GR data up to the surface through the casing while pulling out of the hole (POOH) using reservoir section logging tools. Although GR measurements are very helpful, they are seldom recorded in the top section. This study utilizes artificial intelligence (AI) techniques—specifically artificial neural network (ANN) and adaptive neurofuzzy inference system (ANFIS)—to predict real-time surface hole GR by leveraging drilling parameters and mud weight data. Training on datasets from six wells with a total of 2,100 data set yielded accurate predictions, with the ANN model slightly surpassing ANFIS in performance. The models were and assessed for accuracy using metrics such as root mean square error (RMSE), average absolute percentage error (AAPE), and correlation coefficient (R). These models offer substantial benefits by providing real-time GR data for analysis, thereby eliminating the necessity for post-drilling recordings. This advancement not only saves time and costs associated with logging but also enhances the efficiency of data-driven decision-making during drilling operations. 1. INTRODUCTION Gamma-ray (GR) logs stand as pivotal tools in formation evaluation within the petroleum industry, offering direct insights into geological formations. Equipped with detectors that measure the total gamma ray intensity, GR logging tools serve various purposes, notably in distinguishing between shale and non-shale formations. This distinction arises from the distinctive response exhibited by shale formations, correlating closely with the shale content within the formation. Consequently, formations rich in radioactive materials like carbonates and sandstone tend to yield lower GR values, whereas increased shale content results in higher GR values (A. A. A. Mahmoud et al., 2017; K. A. Mahmoud et al., 2019; Wang et al., 2018). Beyond lithology identification and zone correlation, GR logs find utility in shale volume calculations within reservoir rocks, facilitating the precise classification of productive and non-productive zones based on predetermined shale volume thresholds (Ibrahim & Elkatatny, 2022). While GR logs serve as valuable inputs for modeling lithological effects, it's notable that they are not typically recorded during surface hole drilling. Instead, the conventional approach involves recording GR data up to the surface through casing during the pull-out-of-hole (POOH) process using reservoir section logging tools. This underscores the necessity for a real-time model capable of lithology identification by predicting GR logs from surface drilling parameters in addition to the mud parameters.