Injection-Induced Seismic Risk Management Using Machine Learning Methodology – A Perspective Study

Miao He,Qi Li,Xiaying Li

doi:10.3389/feart.2020.00227

Abstract

Effective identification of induced seismicity and real-time management of seismic risks are hot topics due to increasing induced seismicity in areas related to energy exploitation. Existing decision-making tool for managing seismic risks, known as the traffic light system, is not robust enough. To meet the increasing needs for safe mining of energy at production sites, finding an advanced and efficient method to improve the traffic light system is essential. In recent years, machine learning, an advanced inductive and analytical method, has been widely used in seismology. In this context, research gaps associated with the identification and management of induced seismicity, as well as the current achievements of machine learning in addressing induced seismicity problems, are reviewed. A basic framework of using machine learning method to optimize the traffic light system in the industrial production process is first proposed. Then, its feasibility and rationality are demonstrated by similar cases. This framework may provide a reference for the development of a risk-based adaptive traffic light management system.

Highlights

Earthquakes can cause the sudden release of elastic energy in the Earth
Managing the risks of induced seismicity is of great significance for the smooth progress of industrial production and the development of energy technology
This method did not seek waveform parameters of detected signals and could classify rock fracture and blast signals automatically based on the self-learning capacity of back propagation neural networks (BPNNs). On this basis, (Zhou et al, 2019a,b, 2020) further proposed an improved joint method based on discrete wavelet transform, modified energy ratio and Akaike information criterion (AIC) pickers, which effectively overcame the interference of spike noise and channel crosstalk and greatly improved the accuracy of automatic method for picking onset time of acoustic emission (AE) signals. These results indicate that using Machine learning (ML) method can capture more comprehensive and accurate rock fracture information at the laboratory scale, which can help us deepen understanding of the mechanism and further improve the physical and statistical models of induce seismicity

Summary

INTRODUCTION

Earthquakes can cause the sudden release of elastic energy in the Earth. Generally, natural earthquakes are caused by crustal movements, and the majority of strong seismic events in the world are natural earthquakes and have tectonic origin. The traffic light system (TLS) has been used to manage the risk of induced seismicity, there remain two unresolved problems at present: (1) the differences between natural and induced earthquakes cannot be completely distinguished; (2) the evolving risk of deep fluid injection-induced seismicity is still difficult to manage effectively and timely, especially in the post-injection phase after shut-in To solve these two problems, many different types of data related to injection-induced seismicity are needed, including geological data, seismicity and operational parameters (Yang et al, 2017). The first section briefly reviews the research progress of induced earthquakes related to fluid injection, including the explanation of the mechanism, the discrimination of induced seismicity and the concept of the traffic light system. Physics-based probabilistic models have been developed, taking into account physical and statistical-stochastic factors (Dahm et al, 2015) This model can be used to quantify the probability of event rate change induced by stress changes.

Methods

CONCLUSION

Findings

Background