A Predictive Model for Unplanned Well Down in Heavy Oil Operations to Support Operational Decision-Making

Abstract

The decline in Heavy Oil (HO) production at PT XYZ is strongly influenced by unplanned well down events, which generate significant Loss Production Opportunity (LPO) and disrupt the achievement of production targets. This study aims to develop and compare two time series forecasting methods, ARIMA and Holt‑Winters Exponential Smoothing (HWES), to predict future well down incidents caused by Mechanical Pumping Unit (MPU) failures. Model accuracy was evaluated using the Mean Absolute Percentage Error (MAPE), and the Seasonal ARIMA (2,2,1) Model was identified as the most accurate, achieving a MAPE value of 4.56 percent, significantly outperforming both HWES variants, which produced much higher errors (highest MAPE 27.37 percent). Using this model, the estimated financial loss in the Base Case scenario is projected at Rp 13.35 billion per year, with the worst sase scenario potentially reaching Rp 41.80 billion. The forecasting results provide substantial managerial value by supporting informed operational decision‑making. Three key strategic implications are obtained. First, financial risk control can be strengthened by using the Upper Bound 95 percent as a basis for justifying MPU upgrade budgets. Second, production target planning becomes more realistic by incorporating predicted LPO values. Third, integrating LPO‑based thresholds into KPI monitoring establishes an early warning system that shifts operational control from reactive to anticipatory.