Bottomhole Pressure Prediction of Carbonate Reservoirs Using XGBoost

Abstract

The bottomhole pressure is one of the key parameters for oilfield development and decision-making. However, due to factors such as cost and equipment failure, bottomhole pressure data is often lacking. In this paper, we established a GA-XGBoost model to predict the bottomhole pressure in carbonate reservoirs. Firstly, a total of 413 datasets, including daily oil production, daily water production, daily gas production, daily liquid production, daily gas injection rate, gas–oil ratio, and bottomhole pressure, were collected from 14 wells through numerical simulation. The production data were then subjected to standardized preprocessing and dimensionality reduction using a principal component analysis. The data were then split into training, testing, and validation sets with a ratio of 7:2:1. A prediction model for the bottomhole pressure in carbonate reservoirs based on XGBoost was developed. The model parameters were optimized using a genetic algorithm, and the average adjusted R-squared score from the cross-validation was used as the optimization metric. The model achieved an adjusted R-squared score of 0.99 and a root-mean-square error of 0.0015 on the training set, an adjusted R-squared score of 0.84 and a root-mean-square error of 0.0564 on the testing set, and an adjusted R-squared score of 0.69 and a root-mean-square error of 0.0721 on the validation set. The results demonstrated that in the case of fewer data variables, the GA-XGBoost model had a high accuracy and good generalization performance, and its performance was superior to other models. Through this method, it is possible to quickly predict the bottomhole pressure data of carbonate rocks while saving measurement costs.