Data Preparation for A Supervised Deep Learning Approach to Detect Microseismic Events using Distributed Acoustic Sensing

Distributed Acoustic Sensing (DAS) is an enabling technology for efficient seismic data acquisition for monitoring of passive micro-seismic events using permanent downhole installations. However, acquiring substantial amounts of data challenges existing computational systems and algorithms, especially for continuous passive seismic monitoring applications. Thus, more than ever, we would require novel methods to analyse such big data. This abstract explores preparing and using modelled data to train machine learning models and address the gap between modelled and field data. We then investigate a supervised deep learning approach to detect and locate microseismic events resulting from CO2 injection. We identified the main challenges of using modelled data to train the neural network and addressed them to fill the gap in the context of a microseismic application. We demonstrated the methodology using synthetic data and evaluated it using the Otway CO2 injection site data. Moreover, we performed more tests to confirm the observed effects of including time shifts in the training data. Those enlightening results pave the way for a more extensive study and potential applications to more field data.