Physics Guided Data Driven Model to Forecast Production Rates in Liquid Wells

Abstract

AbstractIn the development of shale plays, significant emphasis has been laid on forecasting well performance based on rates and finding the expected ultimate recoveries. Specifically, forecasting producing gas-oil-ratio (GOR) over the long term has been problematic, given the complexities and uncertainties in modeling a muti-stage fractured horizontal well in the unconventional reservoir. In this work, we propose a hybrid model which is capable of accurately forecasting multiphase flow rates. The proposed hybrid forecasting modeling is an amalgamation of data-centric methodology blended with physics-based principles, using easily available inputs such as production rates, flowing pressure, and fluid properties. The proposed method is a two-step procedure – (1) detect the inflection point up to which the gas produced is only the solution gas using an automated trajectory detection procedure, imposing physics-based constraints (2), apply the material balance to calculate dynamic drainage volume, average reservoir pressure, and productivity index that are used to forecast well performance in the future. The proposed approach also handles changing artificial lift strategies and hence changing bottom hole pressure conditions, which is a practical consideration since most unconventional wells experience operational changes throughout their lifecycle. The automated trajectory detection procedure consistently captures the inflection point for all wells and is robust to scale for all well types. The history-matched multiphase flow model parameters are blind-tested to validate the model. The proposed technique extrapolates reservoir pressure depletion based on established trends to forecast GOR trends with reasonable accuracy at an extremely low computational cost. The proposed hybrid model overcomes (1) deficiencies of pure data-driven approaches, where changes in operating conditions are not properly represented and the forecasts are not physically consistent, (2) limitations of analytical models, where the assumptions are too many/strict to represent the real-life performance of a multifracture horizontal well, and (3) complexities of numerical simulation models, which are expensive, time-consuming and requires too many inputs for initialization. Additionally, the proposed hybrid model provides a robust and scalable method to identify future GOR trends to support the pace of operations and data-driven decision-making.