Machine Learning Algorithms Comparison for Stuck Pipe Detection in Geothermal Drilling Operation in North Sumatera, East Java and West Java Fields

Abstract

Intensive geothermal drilling efforts are essential in order to reach the Indonesian government target of 3,355 Mwe by 2030. To reduce the overall risk of stuck pipe, a significant risk in drilling geothermal wells, innovations such as Stuck Pipe Machine Learning (SPML) were developed to support engineers' ability to reduce potential problems during drilling and optimize the overall drilling operations. This research aims to find the best machine learning algorithm for stuck pipe detection in geothermal drilling operations and how the results would be implemented using historical data. Literature studies were conducted prior to the development of machine learning with 15 wells of data from three geothermal fields in Indonesia. Ten drilling parameters were selected, cleaned, preprocessed, and prepared for use in four machine learning algorithms (K-Nearest Neighbors, Random Forest, Decision Tree, and Multi-Layer Perceptron). These four algorithms were trained and tested with 12 geothermal wells, with 70% of the data for training and 30% of the data for testing. Then it was also tested using historical data (blind tests) for three geothermal wells. After the comparison, it was found that Multi-Layer Perceptron is the best algorithm, with the accuracy on the blind test of 3 geothermal wells being 0.97, 0.94, and 0.99, respectively. This overall accuracy is very, very good, and in the evaluation, it is also true that it can detect normal, pre-stuck, and stuck pipe conditions well. Finally, Stuck Pipe Machine Learning (SPML) will continue to be developed for the purpose of geothermal drilling optimization in Indonesia.