Seismic facies analysis from well logs based on supervised classification scheme with different machine learning techniques

摘要

Seismic facies analysis (SFA) has been proven to be useful in interpreting seismic data, allowing significant information about subsurface geological structures to be extracted. SFA uses different seismic attributes extracted from 3D seismic data and well logs to construct detailed seismic facies and lithology variation maps. Accurate seismic facies map could be used for reservoir exploration, optimization, management, and development. The recent SFA research was mainly based on unsupervised methods, with few works done using supervised classification. Therefore, this study focuses on supervised classification for qualitative mapping of the reservoir facies distribution. Precisely five well-known classifiers, called the multilayer perceptrons (MLPs), support vector classifier (SVC), Fisher, Parzen, and K-nearest neighbor (KNN). Each classifier was tested to provide an opportunity for the direct assessment of their feasibility in the classification of facies. The approach was applied on the carbonate reservoir from a real oil field in Iran. The numerical relative errors associated with different classifiers as a proxy for robustness of SFA provides reliable interpretations. Our results show stability of both SVC and MLP classifiers in supervised-classification-based studies although SVC has relatively better results.