APPLICATION MACHINE LEARNING (DECISION TREE) IN DETERMINING REACTIVATION CANDIDATES IDLE WELL AT PT. PERTAMINA EP REGIONAL 4 ZONE 11 CEPU FIELD
Boni Swadesi, Hariyadi, Damar Nandiwardhana, Geovanny Branchiny Imasuly, and Herlina Jayadianti

UPN Veteran Yogyakarta

Abstract

Indonesia has a tough challenge with realizing oil production of 1 million barrels per day by 2030 by relying on old fields or mature (brownfield), which seeks to exploit the remaining hydrocarbons. One of the targets is the reactivation of idle wells at PT. Pertamina EP Regional 4 zone 11 Cepu field, research and innovation development focused on production is carried out idle well.
In this research, reactivation candidates were determined to idle well. The first stage is to define the problem to understand the influencing factors of idle well and information about recent developments in predicting the determination of reactivation candidates. Data collection in the form of primary and secondary documents for 2018-2023. The next stage is implementing Machine Learning (ML) (Decision Tree (DT)) to be able to overcome the problems of data accuracy and complexity, as well as create efficient and accurate classification patterns, a Web Application which can help decision-makers in determining which wells should be reactivated which can provide the best solution to the problem of increasing oil recovery.
The research results show a high success rate on Accuracy Under Curve (AUC) and Receiver Operating Curve (ROC), amounting to 0.99 which shows that the classification model has a high probability, using entropy two potential wells were obtained for reactivation based on Lifting method (ESP) namely wells NGL-P-001 and TPN-004 with Well Cum Prod (Np, MBO) (107.89 then 132.570 MBO) then HC Remaining Potential (Oil, MBO) (4,609 and 52.42 MBO), with recommendations for improvement, namely Well Service. Of the two wells that meet the reactivation criteria based on the model decision tree, an evaluation is carried out by Decline Curve Analysis (DCA) where the lower the MSE, the better the fit and model Hyperbolic and Stretched Exponential, which yielded the lowest values of 1106.6 and 1142.35, thus indicating that the model may be the best fit among the models considered. The results of forecast production rate vs cumulative oil production were used to predict future oil production and cumulative oil amounting to 4451.22 BBL was obtained.

Keywords: Machine Learning, Reactivation, Idle Well, Decision Tree, Increased Oil Recovery

Topic: Engineering