Machine Learning Based Decline Curve Analysis (DCA) for Forecasting Crude Oil Production in a Mature Field in Nigeria

Abstract Mature Oil fields constitute a significant percentage of Nigeria's total Hydrocarbon production with a high concentration in the Niger delta region. Accurate forecast of the production rates in these fields is crucial for efficient resource and operational management. This study investigates application of Machine Learning (ML) techniques to improve decline curve analysis (DCA) for production forecasting in Nigerian mature oil fields with emphasis on waterflooded reservoirs in the Niger Delta. A single well from a water drive marginal field was analysed using historical production data. Three (3) different ML models including Gradient Boosting (GB), Random Forest (RF) and Support Vector Regression (SVR) were adopted for this study and compared with results from conventional DCA. Random Forest Model emerged as the best predictive model for the dataset achieving a mean absolute error of 5.0028, a root mean squared error of 17.7029, and R2 value of 0.9939. Model evaluation further indicated that a production rate threshold of 200 barrels per day and a water cut limit of 40% are critical parameters influencing production decline behaviour. The findings confirm that ML approaches, particularly ensemble learning models like Random Forest can significantly enhance the accuracy of production rate forecasting over conventional DCA. This reserch contibutes to a deeper therotical understanding of predictive modeling in petroleum engineering and offers pratical insights for optimizing field devlopment startegies in mature, waterflooded reservoirs of the Niger Delta.