SUN Zhixue^1,2, JIANG Baosheng¹, XIAO Kang³, LI Jikang¹

1. School of Petroleum Engineering, China University of Petroleum (East China), Qingdao 266580, Shandong Province, China;

2. Key Laboratory of Unconventional Oil & Gas Development, Ministry of Education, China University of Petroleum (East China), Qingdao 266580, Shandong Province, China;

3. African Research Institute, PetroChina Research Institute of Petroleum Exploration & Development, Beijing 100083, China

Abstract: The natural fractures are important for oil and gas storage and transportation in the bedrock buried hill reservoirs. The fracture aperture is the key parameter for the reservoir quality characterization as well as reserves and productivity evaluation of buried hill reservoirs. In this study, a novel fracture aperture prediction algorithm is proposed based on the ensemble learning algorithm. The samples are collected from the bedrock buried hill reservoirs in Basin B of Chad, Central Africa, and their fracture aperture data are extracted from the sample description, key well imaging logging, and fracture parameter interpretation. The logging data at the same depth are used as the feature variables to constitute the learning sample, and the K-means clustering algorithm is applied to reducing noise of the learning sample and eliminate abnormal data. Through support vector machine (SVM) regression and XGBoost regression algorithm, and by using random search that optimizes model parameters, the fracture apertures are estimated according to the basic models combined with ridge regression. The results show that the performance of the novel ensemble learning algorithm is better than that of the basic model; the root mean square error between the predicted and actual values of the test set is 0.047, and the correlation coefficient is 0.931. The algorithm mitigates the instability of the single regression algorithm, improves the generalization ability, and provides a new way to aperture prediction.