Production Estimation for Shale Wells with Sentiment-Based Features from Geology Reports

Shale oil and gas have become very promising unconventional energies in recent years. To optimize operations in oil and gas production, a reservoir model is important for understanding the subsurface appropriately. Generally, sensor data, such as surface seismic data, are most popular data sources in modeling the reservoir with either a numerical simulation model or an Artificial Intelligence (AI)-based model. In this paper, to obtain data that describe the subsurface more exactly, information, including phrases that indicates possible bearing oil or gas and rock colors, is extracted from geology reports. Sentiments of the phrases is identified by sentiment analysis, and sentiment sequence over measured depths is then used to generate features. The rock-color similarities between wells are calculated as well, and integrated as distance metrics into a geology-based regression method. Extensive experiments on Bakken wells in the United States show the effectiveness of using the features extracted from geology reports and the rock colors in terms of estimating well production.