(582c) Integrating Public and Private Data for Modeling and Optimization of Shale Oil and Gas Production

The overarching goal of this project was to develop novel methods to facilitate rapid data transformation into knowledge, for subsurface applications. In the first phase of this effort, the project team developed workflows and interactive applications that unite data from a variety of public and private sources in machine-readable data formats to drive machine learning (ML) and artificial intelligence (AI) applications, including methods to assess data quality, identify data gaps, fill data gaps through science-based approaches, and demonstrate that science-informed AI/ML can be used for near real-time scenario analysis and optimization of field development planning and resource recovery. The study implemented a data-intensive supervised machine learning approach, through a quasi-experimental framework, with the objective of quantifying the impact of oil and gas operator-specific proprietary data on ML-based predictive model performance as compared to using oil and gas datasets that are publicly available. Using the time-series data, the long short-term memory (LSTM) based neural network models were designed and trained to jointly predict daily oil, gas, and water production for horizontal wells as a function of bottom-hole pressure drawdown, spatial placement across the study domain, and well completion attributes. The combination of input features was set to align with either one of three categories: 1) predominantly proprietary in origin (proprietary model); 2) a mix of data of proprietary and public origin (quasi-public model); or 3) only the data commonly available from public or vendor sources (public model). The best-performing model utilized the proprietary data for downhole pump inlet pressure as a key input attribute to simulate the hydrocarbons production response to various pressure management scenarios. The study specifically examined utility of the proprietary deep learning model developed by the team in forecasting unconventional oil and gas production by using well data from the Permian Basin (Midland sub-basin, Texas, United States). The model predicted daily production for oil, water, and gas with accuracy on the order of 79 percent (for water and gas) to 86 percent (for oil). The extensive input parameter set used in the development of the model provides the utility to test and evaluate multiple controlling features on associated production. These features include (but are not limited to) geologic properties from well log data, artificial lift design data, well placement (spatial, depth, and wellbore trajectory orientation) and completion attributes, detailed well hydraulic fracturing data, well operating conditions, and operational controls of production intensity (via pumping pressure downhole). Results of the scenario planning exercise indicated that rapid drawdown of pressure generated the initially higher oil and gas production. However, the cumulative volumes of oil and gas production in response to rapid drawdown strategies were lower compared to conservative drawdown strategies that sustained pressure over the medium-to-long term. The rapid drawdown strategies also resulted in lower water production over the lifespan of the well compared to conservative drawdown strategies. This production forecast of the varying drawdown strategies may have significant operational and economic implications, with contrasting perspectives between well productivity and profitability given the typical oil and gas economics and the volatility in the oil and gas markets. Further analysis using the proprietary ‘shut-in’ and ‘frac-hit’ data provided the means to estimate the potential oil production lost due to ‘frac-hit’ events at 287 wells in the proprietary dataset made available by the project partners. The historical oil price ($/bbl) was correlated with the dates of ‘shut-in’ and ‘frac-hit’ periods. For the entire field, the cumulative sum of all ‘lost-production’ events returned an estimated annualized loss of approximately $30 million.