Application of Xgboost Algorithm Based on Machine Learning in Reservoir Prediction of Offshore Oilfield

Conventional seismic reservoir prediction usually faces problems such as low signal-to-noise ratio, low resolution, and long research period. It is difficult to efficiently implement fine reservoir prediction, especially for thin reservoir. Therefore, machine learning technology is applied to high-precision reservoir prediction, with high-dimensional, non-linear characterization of reservoir characteristics through multiple seismic attributes. Based on seismic, well logging, geology and other datas of M oilfield in South China Sea, they are collected and preprocessed firstly. The preferred sesmic attributes and logging datas are encoded and structured as label datas. Taking the geological model as the unified output carrier, different types of datas are outputed based on the same dimension, so as to obtain a large number of machine learning samples. The samples are divided into training sets and validation sets for cross validation training. The comparison and error analysis of various machine learning algorithms are carried out. The optimal algorithm is used to establish the reservoir prediction model and the blind wells are used for verification. The results show that the ensemble algorithm has an advantage over the single machine learning algorithm in reservoir prediction, and the XGBoost model has the best prediction accuracy and stability among the several mainstream machine learning algorithms. Its reservoir prediction rate with the wells is more than 85%, and the vertical resolution reaches 2–3 m. The accuracy of Random Forest is lower than XGBoost model, but higher than Decision Tree and Support Vector Machine. Compared with conventional reservoir prediction methods, the machine learning method is faster and more efficient in calculation, better coincidence with the actual drilling data, and less dependent on the experience of researchers. The machine learning method improves the accuracy of thin reservoir prediction in M oilfield and provides a direction for oilfield exploration in later stage, which offers a new technical means for reservoir prediction in other oilfields.