Application of Machine Learning Algorithm Approach to Enhance the Workover Program Success Ratio in Pertamina Hulu Sanga Sanga Field, Sanga-Sanga Block of East Kalimantan, Indonesia

Abstract The workover program significantly contributes to the Pertamina Hulu Sanga Sanga field's total production. Each year, the number of perforation jobs increases due to an increase in new wells. However, due to organizational restructuring, the number of engineers available to analyze these candidates has decreased, leading to a decline in the quality of workover outcomes. To tackle this issue, implementing machine learning aims to maintain the massive of workover jobs while ensuring the accuracy of perforation candidate results. This study utilizes 12,247 cleansed historical workover data points incorporating data attributes such as completion data, cumulative production from the same reservoir zone, and reservoir properties. To ensure accuracy, the dataset is split into an 80:20 ratio for training and validation purposes. The Decision Tree machine learning algorithm is selected for the analysis, which provides FLOW or NO FLOW labels, signifying successful oil or gas extraction, respectively. Once validated, the model predicts perforation outcomes in the Mutiara, Pamaguan, Badak, Nilam, Lampake, and Semberah Field, Sanga-Sanga Block, contributing to the advancement of predictive techniques in the field. After training on the available data, the machine learning model achieved an impressive accuracy of 77.5%. The model has since been utilized to predict the outcomes of 13,791 remaining unperforated reservoir, and these results are available to engineers through an interactive dashboard. This dashboard enables engineers to identify perforation candidates instantly and accurately by providing predictions of FLOW or NO FLOW in the desired reservoir. Moreover, a notification system notifies the engineers of the perforation candidates every month, based on specific criteria. To monitor models, each new workover task is compared to the model's prediction of FLOW or NO FLOW result. At present, the model has an accuracy rate of 91.5%, highlighting the efficiency of machine learning in providing a high-accuracy list of workover candidates. Compared to conventional analysis, this approach saves significant time and reduces workload. With an enhanced workover success rate, this method substantially elevates the success ratio of workover despite reservoir engineers contending with heightened workover demands and an increased well count. Integrating Perforation Machine Learning (PerfoML) into the traditional quality review process has not only improved the quality and quantity of workovers but has also helped Pertamina Hulu Sanga Sanga exceed the production realization target by over 110% as set in the KPI target for 2022-2023. A breakthrough in engineering has been discovered through the utilization of non-conventional and highly abundant historical data by employing simple machine learning algorithms. This digitalization initiative has led to significant economic benefits in terms of production and cost efficiency.