Identifying Discrete Fracture Networks by Clustering with Microseismic Data

ABSTRACT: Microseismic monitoring is common in the mining and petroleum industries. Pattern recognition to identify faults, fractures, and damage zones in the microseismic data can be a challenge due to the density and number of observations and their multiple attributes. This study applies unsupervised machine learning techniques to identify features in a dataset from a hydraulic fracturing program in the Duvernay Formation. It compares partitional, hierarchical, and density based clustering methods using various validation statistics and with multiple subsets of data attributes. The study shows that lower dimensional datasets tend to yield the best results, and that it is possible to cluster microseismic data using unsupervised techniques.