Improving Crude Oil Supply Planning using Forecasing and Digital Twin-Decision Support: A Case Study at Refinary Company

PT. X is a Refinery Company with core activities focused on processing crude oil into fuel, non-fuel, and petrochemical products. Effective planning of crude oil demand and supply is therefore essential to ensure operational continuity and national fuel availability. This study focuses on two categories of crude oil feedstock used, namely heavy crude and super heavy crude. The objective of this research is to evaluate discrepancies between planned and realized crude oil orders and to generate forecasts for the subsequent 12 periods to support crude oil supply planning. Under existing conditions, crude oil supply scheduling relies on Linear Programming (LP) tools as input for the monthly Crude Master Program meeting. The initial Mean Absolute Percentage Error (MAPE) values were recorded at approximately 34% for heavy crude and 8% for super heavy crude, indicating deviations between planning and realization. To improve forecasting accuracy, this study applies a seasonal multiplicative decomposition method to time-series forecasting. The results demonstrate that the proposed method reduces the MAPE values to approximately 18% for heavy crude and 7% for super heavy crude, indicating a significant reduction in planning deviation. Beyond improved numerical accuracy, the forecasting results provide predictive insights that support early detection of supply deviations and proactive decision-making in crude oil supply planning. Furthermore, this study positions forecasting as a predictive intelligence component within a Digital Twin–ready decision support framework, contributing to more intelligent, data-driven, and coordinated crude oil supply management.PT. X是一家炼油企业，核心业务聚焦于将原油加工为燃料、非燃料及石化产品。因此，有效规划原油供需对保障运营连续性与国家燃料供应至关重要。本研究聚焦于所使用的两类原油原料：重质原油（heavy crude）与超重型原油（super heavy crude）。本研究的目标为评估计划与实际原油订单之间的偏差，并生成未来12个周期的预测结果，以支撑原油供应规划工作。在现有条件下，原油供应调度依赖线性规划（Linear Programming，LP）工具作为月度原油总计划会议的输入依据。初始平均绝对百分比误差（Mean Absolute Percentage Error，MAPE）值显示，重质原油约为34%，超重型原油约为8%，表明计划与实际执行存在偏差。为提升预测精度，本研究将季节性乘法分解法应用于时间序列预测。结果表明，所提出的方法将重质原油的MAPE降至约18%，超重型原油的MAPE降至约7%，规划偏差得到显著降低。除数值精度提升外，预测结果还提供了预测性洞察，可助力提前发现供应偏差，并在原油供应规划中开展前瞻性决策。此外，本研究将预测定位为适配数字孪生（Digital Twin）的决策支持框架中的预测智能组件，有助于实现更智能、数据驱动且协同一致的原油供应管理。