Knowledge-Based Machine Learning Approaches to Predict Oil Production Rate in the Oil Reservoir

Predicting the oil production rate is a crucial means to improve the operation of hydrocarbon reservoirs and manage the economic plans for oil companies. However, developing a reliable model to predict oil production rate using traditional numerical frameworks are challenging and requires too much time to attain a single model. Thus, in this paper, Machine Learning (ML) techniques are presented as a robust and intelligent framework to predict oil production rates accurately and timely. The ML techniques include Multiple Linear Regression (MLR), Random Forest (RF), Decision Tree (DT), and K-nearest neighbor (KNN). These four techniques were engaged to predict the oil production rate of real oilfield data of 11 wells. The 11 oil wells were considered as datasets to achieve a precise prediction of the oil production rate. The available datasets were split into two subsets of training and testing data sets. Furthermore, the Root Mean Squared Errors (RMSE) and determination-coefficient values (R2) regression metrics were employed to evaluate the model performance. Hence, the comparative analysis of the proposed models was presented for all 11 selected production wells. The analysis of results showed that RF can be considered the best predictive ML model for predicting oil production rate with the lowest RMSE and the highest R2 scores in all 11 production wells. In the KT911H well, the RF model achieved the most accurate results with RMSE and R2 0.868 and 0.9993 respectively. In addition, the study analysis illustrated that the RF can be considered to be the best predictive model, which was also applied to indicate the relationship between the input parameters and the oil production rate. A sensitivity analysis of the RF model indicated that the liquid volume, water cut, and gas pressure are the most important input parameters affecting the oil production rate performance in all 11 production wells. This paper therefore presents a pragmatic approach for predicting the oil production rate of a typical oilfield and the parameters with the most effects on the prediction based on machine learning techniques.