A Novel Approach to Discovery of Hidden Structures in Microseismic Data Using Machine Learning Techniques
Conference Paper
Overview
Identity
Other
View All
Overview
abstract
AbstractThe objective of this paper is to propose an alternative data analysis approach to working with microseismic data. Modern machine learning techniques, such as MWCA (Multiway Component Analysis) and TD (Tucker Decomposition) can give the capability to efficiently work with complex high-dimensional microseismic data structures. Using this method, it was possible to restore hidden information about the signal, compress the data, and get insights about fractures without using conventional time-consuming simulations. Therefore, it is an important addition to the hydraulic fracturing quality assessment. It is a cost-effective technique providing a greater degree of automation in comparison to conventional methods.The approach was tested on synthetic data and relevant real microseismic data provided by a service company. The data was integrated in a 3rd-order tensor form where modes are: seismic events time, receiver locations, and event locations. The tensor was then decomposed into a core tensor and three factor matrices by means of a special form of TD called HOSVD (Higher-Order Singular Value Decomposition). HOSVD is a multidimensional decomposition used to extract low rank approximations of tensors. The MWCA technique was utilized to impose constraints on TD. HOSVD showed potential as a tool for a rapid fractures analysis by observing decomposed tensor structure. Additionally, the technique helped reduce the original model by 73% (supercompression).The proposed workflow is general and highly applicable to various plays. Since the applications of MWCA and TD are still emerging, future enhancements to this methodology are expected. In turn, this will reveal further insights from microseismic data, making it paramount to optimal fracturing and improved field management.
