Multiple Linear Regression Models for Crude Oil Forecasting in the A-NC115 Field, Libya

Abstract

Accurate multi-horizon production forecasting remains a critical challenge in mature Libyan oilfields. In this paper, a Multiple Linear Regression (MLR) approach is proposed to forecast crude oil production in the A-NC115 oilfield, Libya, by identifying the main influencing variables. Unlike most previous studies conducted in Nigerian and other international fields, this study applies MLR forecasting to a mature Libyan oilfield, addressing an important regional research gap. The A-NC115 field is a mature water-supported reservoir where accurate forecasting of oil production is of essential importance. The trends of production, pressure, and water injection directly impact the operational and development decisions. the three different models were developed and evaluated: a short-term model based on daily data, a mid-term model based on monthly data and a long-term model based on yearly data. Historical production data from February 2018 to December 2023 was collected, pre-processed and analyzed using Minitab software. Model performance was evaluated on a hold-out testing dataset (Jan 2024 - Apr 2024) using statistical metrics such as R-squared (R2), Adjusted R2, Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The best accuracy was achieved for the daily model (R2= 99.99%, MAPE = 7.20%) followed by the yearly model (R2= 99.84%, MAPE = 8.67%). The monthly model is still quite explanatory (R2= 99.96%) but with a higher MAPE (17.87%) showing a higher variance of prediction at this specific temporal scale. The results suggest that data-driven models provide accurate forecasts that support improved resource management, maintenance planning, and more efficient operational and strategic decision-making, along with a better understanding of future needs such as equipment upgrades, associated water management, and production optimization.