Thickness prediction for high-resolution stratigraphic interpretation by fusing seismic attributes of target and neighboring zones with an SVR algorithm. (March 2020)

• Record Type: Journal Article
• Authors: Li, Wei; Yue, Dali; Wu, Shenghe; Shu, Qinglin; Wang, Wenfeng; Long, Tao; Zhang, Benhua
• Is Part Of: Marine and petroleum geology. Volume 113(2020)
• Publication Date: 2020-03
• Subjects: Sand thickness -- Seismic attribute -- Seismic forward model -- Neighboring zones -- Machine learning -- Support machine regression; Submarine geology -- Periodicals; Petroleum -- Geology -- Periodicals; Géologie sous-marine -- Périodiques; Pétrole -- Géologie -- Périodiques; Petroleum -- Geology; Submarine geology; Periodicals; Electronic journals; 551.468
• DOI: 10.1016/j.marpetgeo.2019.104153
• Languages: English
• ISSNs: 0264-8172

Abstract

Abstract: Predicting the thickness of fluvial sandbodies is significant for hydrocarbon exploration and development programs. Owing to spatially dense sampling, the analysis of sand thickness based on 3D seismic data has become one of the most popular methods. However, in terms of high-resolution stratigraphic interpretations, the thickness range of the target zone is usually less than the seismic temporal resolution; so seismic responses in the target zone are significantly affected by the responses of the upper and lower zones (neighboring zones). Therefore, we developed a new method of sand thickness prediction to reduce the interferences of neighboring zones by fusing the seismic attributes of the target and neighboring zones based on machine learning with a support vector regression (SVR) algorithm. First, the sand thickness values interpreted from wells were set as supervised data, and seismic attributes (around the wells) in the target and neighboring zones were input as training data. Then, an SVR model was trained using both sets of input data. Finally, the attributes in the target and neighboring zones were inverted into the predicted sand thickness using the trained SVR model. To test the proposed method, a multi-thin-bed, 2D model was designed, and a complex, geologically realistic, 3D model was established based on well-log-based facies interpretation using an object-based modeling method. This sand-thickness prediction method was also applied to a real seismic dataset of Chengdao Oilfield in the Bohai Bay Basin of China. These applications demonstrate that the proposed method can significantly reduce the interference of neighboring zones and improve sand thickness prediction. Highlights: Seismic responses in the target zone are usually affected by neighboring zones. A new method of fusing attributes in the target and neighboring zones was proposed. The new method can markedly reduce the interference of neighboring zones. The proposed method distinctly improved the prediction of sand thickness.