Hydrocarbon reservoir prediction using support vector machines

Hydrocarbon reservoir prediction using seismic features is a typical classification problem. Numerous methods have been developed for computer-aided reservoir prediction. The prediction accuracy is restricted by the following facts: 1) small amount of samples; 2) small size of features; and 3) the intricate non-linear relation between features and reservoir level. This paper proposes a feature expansion and feature selection method, which maps the features to a higher dimensional feature space and then select proper features, thus mines the 'true' features. The selected features are used for training a linear classifier. Test with seismic data from Guanyinchang district of Sichuan Province and Chengdao district of Shandong Province, the proposed method achieved better prediction result than other methods.