Introducing attentive neural networks into unconventional oil and gas violation analysis and emergency response system

With the prosperous development of unconventional oil and gas (UOG) began in the mid-1990 s, the proliferation of digital textual compliance reports from the UOG production life-cycle makes it imperative for experts to develop efficient ways of supporting emergency responses based on the textual based data sources. In this respect, we utilized the UOG compliance reports from the Pennsylvania Department of Environmental Protection from 2000 to 2019, then established an attentive neural-network framework to support on-site emergency responses. The advantages of attentive-based neural networks over the other mechanisms are that it not only generates powerful contextual vectors for follow-up tasks but also it allows us to observe the importance of violation factors with respect to different scenarios. The experimental results show that our model can extract valid representation from narrative texts in UOG violation compliance reports and achieve high performance in emergency response. At the same time, we obtained two intriguing practical implications: first, geographical and time characteristics are powerful indicators for supporting decision making in UOG on-site emergency responses; second, there is an urgent need for governments to implement different inspection strategies according to unique UOG sites rather than counties concerning specific geological features, which benefits from saving human labor and financial expenditures.