Application of machine learning for evaluating and predicting fault seals: A case study in the Huimin depression, Bohai Bay Basin, Eastern China

Abstract

Fault seal is of great significance for hydrocarbon migration, accumulation, and further hydrocarbon reservoir production. Approximate 32% petroleum resource is confirmed to be related to the faults. However, the existing fault seal evaluation methods based on statistical analysis cannot accurately predict fault seal which is influenced by multiple factors in a complex way. It is necessary to improve the fault seal evaluation methods for enhancing the exploration success rate. In this study, a new fault seal evaluation and prediction method based on decision tree (DT) and random forest (RF) is introduced. First, the original dataset was set by quantification and feature engineering work. Second, the nonlinear classification models for fault seal evaluation and prediction using a binary decision tree named the classification and regression tree (CART) were constructed and improved by overfitting calibration. Third, the random forest algorithm was selected as an ensemble learning method to improve the faut seal evaluation and prediction accuracy. Third, the evaluation metrics and cross-validation were used to evaluate the performance of the model. Finally, the validation test is applied for testing the reliability of the model. The result showed that among the 100,000 models constructed in this study, the DT best model could evaluate and predict the fault seal with a cross-validation accuracy of 80.60% after overfitting calibration by pruning. The best RF model showed the highest test accuracy of 86.54%, which is higher than that of the DT model. The models were used for predicting fault seals in another district in the Huimin Depression, and the prediction accuracy reached 90% and 95% for the DT and RF model, respectively. This study not only introduced a new method for fault seal evaluation and prediction, but also provided guidance for the application and development of machine learning in petroleum exploration and exploitation field and industry.