Machine learning-based production forecast for shale gas in unconventional reservoirs via integration of geological and operational factors

Abstract Hundreds of horizontal wells have been performed fracturing operations to exploit the unconventional shale gas resources in the Duvernay Formation of Fox Creek, Alberta. Despite achieving the practical analysis of shale gas production via the data-mining approach, previous studies failed to incorporate comprehensive site-specific geological and operational factors. In this study, a comprehensive machine-learning approach is developed to forecast the shale gas production via the integration of geological and operational factors. Thirteen geological and operational parameters deriving from the well logging, core experiment and treatment data are included as the input variables, whereas the 12-month shale gas production is regarded as the target variable. Results show that factors that mostly contributed to the shale gas production are found to be total fluid injection, total proppant mass, well TVD, permeability, y coordinate, porosity, gas saturation, number of stages, x coordinate, formation pressure, horizontal length, distance to fault and Duvernay thickness. Four machine learning methods are evaluated, where the Extra Trees approach has led to the highest coefficient of determination R2 of 0.81. Case study for Well 2 have shown that the shale gas production can be doubled if increase the total pumped volume and proppant placed mass by approximately 73% and 38%.