Application of Parallel Support Vector Machine Algorithm for Fault Mapping in 3D Seismic Volume

The data collected for the study of hydrocarbon fields can be massive in terms of volume and variety, such as seismic and well log data, which can be analysed using big data techniques. Quantitative data is gathered from a variety of sources, with different acquisition times and spatial resolutions. In order to obtain information about reservoirs, fault mapping is necessary. Currently, repeated validation and visual examination by specialists is the standard procedure for fault mapping. However, integrating big data analysis tools into the techniques of fault mapping is an attempt to evolve an automated process. Retrieving such vital information from big data will require the employment of scalable data management and algorithms. Parallel algorithms are the most effective for this. Support Vector Machine (SVM), being one of the effective two-class classifiers, is implemented for the mapping of faults. But SVM has the problem of scalability. The parallel SVM (PSVM) plays a crucial role in reducing computing cost. Instead of using the huge quantity of data as the whole set of training vectors, data is split into subsets. Next, SVM is applied separately to the subsets, and then individual results are combined and filtered to get the final fault mapping results. The method has been successfully tested on an offshore 3D seismic dataset from the Krishna-Godavari basin in India.