Novel hybrid artificial neural network based autopicking workflow for passive seismic data

Abstract

Microseismic monitoring is an increasingly common geophysical tool to monitor the changes in the subsurface. Autopicking involving phase arrival detection is a common element in microseismic data processing schemes and is necessary for accurate estimation of event locations as well as other workflows such as tomographic or moment tensor inversion, etc. The quality of first arrival picking is dependent on the actual seismic waveform, which in turn is related to the near surface and subsurface structure, source type, noise conditions, environmental factors, and monitoring array design, etc. We have developed a new hybrid autopicking workflow which makes use of multiple derived attributes from the seismic data and combines them within an artificial neural network framework. An evolutionary algorithm scheme is used as the network training algorithm. The autopicker has been tested and its applicability has been validated using a synthetically modelled seismic source, with promising results. In this work, we share the basic workflow and different attributes that have been tested with this algorithm for a synthetic data set to provide a framework for independent implementation, use and validation. We also compare the results obtained using the new neural network based autopicking routine with very robust contemporary autopicking algorithms in use within the industry.

Automatic Waveform Classification and Arrival Picking Based on Convolutional Neural Network

“…Machine learning methods, supervised classification or unsupervised clustering, are also used in seismic classification and arrival picking (Chen, 2017;Knapmeyer-Endrun & Hammer, 2015;McCormack et al, 1993;Muller et al, 1998;Provost et al, 2017;Rouet-Leduc et al, 2017;Shahnas et al, 2018). The Neural network is one of the mostly used machine learning methods for waveform classification and arrival picking (Akram et al, 2017;Dai & MacBeth, 1995, 1997Langer et al, 2003;Maity et al, 2014;McCormack et al, 1993;Pulli & Dysart, 1990;Zhao & Takano, 1999). Neural networks have powerful learning capabilities without prior knowledge from multiple input features.…”

Novel fracture zone identifier attribute using geophysical and well log data for unconventional reservoirs

“…Microseismic data processing involves multiple steps starting with the use of a simple energy ratio-based triggering scheme to identify potential events in the passive seismic data sets by comparing with predefined normalized thresholds. The triggered data are appropriately time stamped and run through an advanced neural network-based autopicker (Maity et al, 2014). These picks are inverted for location and velocity.…”

A robust neural network-based approach for microseismic event detection

“…In practice, event detection is performed by automatic algorithms which are typically based on some attribute-computation in moving windows (for instance, STA/LTA; Allen, 1982), waveform cross-correlation (Song et al, 2010) or on artificial neural networks (ANNs) (Maity et al, 2014). An ANN based approach use effective models from data-driven learning for solving complex and ill-posed problems (Gentili and Michelini, 2006;Maity et al, 2014). Therefore, it can provide better microseismic event detection as compared to conventional algorithms for low S/N and complex waveforms.…”