Search

The growing maturity of hydraulic fracturing and horizontal drilling techniques has made shale gas production the core energy strategy for many countries. However, it is also bringing along increasing concerns about hydraulic fracturing-induced seismicity (HFIS). In fact, how to effectively assess the seismic risk in hydraulic fracturing areas, identify the key HFIS factors and safeguard people's lives and property has become a scientific problem in need of immediate solution. In this study, a quantitative seismic risk assessment model for hydraulic fracturing areas is built based on rough set and Bayesian network. First, five factors most relevant to HFIS are identified based on existing research experience and the geological and engineering background of the study area. Then, knowledge acquisition on HFIS is performed using the rough set method. Next, the initial Bayesian network is built with the knowledge acquired and a geospatial correction model is constructed by introducing a "latent factor" into it. Finally, the seismic risk level in the study area is quantitatively assessed with the geospatial correction model. Besides, through sensitive analysis of the Bayesian network, it is established that injection volume is the most sensitive risk factor for HFIS and it plays a crucial role in the occurrence of seismic events. Through backward inference of the Bayesian network, the most probable combination of risk factors when the study area is in medium or high seismic risk is identified as injection volume, depth, b value, and fault. Experiments show that our new model is effective; the results obtained can provide scientific basis for predicting seismic risk in hydraulic fracturing areas, improving the hydraulic fracturing construction plans, thereby achieving the goal of resource exploitation and seismic disaster avoidance. • A quantitative seismic risk assessment model is built by combining the merits of both rough set and Bayesian network. • The seismic risk level in the hydraulic fracturing area is assessed for each divided grid. • The degree of influence of individual risk factors on risk events is discussed through sensitivity analysis. • The probable combination of inducing factors is established through backward inference. [ABSTRACT FROM AUTHOR]