A Hybrid Data and Physics Modeling Approach Towards Unconventional Well Performance Analysis

Abstract

Abstract As unconventional reservoirs have become more challenging to develop and predict, understanding well performance has proven to be essential for driving value. Though there have been continuous advancements in well performance analysis and production forecasting for unconventional reservoirs that exhibit prolonged transient flow conditions, there is a still a gaping need for robust and scalable methods which are usable from a practical standpoint, considering data availability and other uncertainties. Traditional forecasting methods (decline curve analysis and its variants) are often not fully representative and impacted by surface operations such as constrained flow, choke changes etc. or subsurface events such as well interference, frac hits, depletion below saturation pressure, etc. Analytical and numerical modeling methods address this issue by applying first principles and simplified physics for integrating flow diagnostics and time-rate-pressure analysis. However, this is often quite interpretive through manual analyses that are not scalable, requiring additional reservoir inputs that are often not collected or known for all wells. Consequently, these methods are not suitable for large scale, repetitive forecasting purposes. We desire a reliable, consistent and scalable well performance analysis method that can work with routinely measured data for most unconventional wells (i.e. surface pressures, flow rates and fluid data). A novel continuous drainage volume estimation method based on the concept of diffusive time of flight is used in conjunction with material balance calculation to estimate pressure depletion and well deliverability from daily flow rates and bottomhole pressure. A data-driven procedure for estimating PVT properties for all wells is also developed and applied to convert surface pressures to bottomhole pressures. A predicted rate profile can be obtained from the productivity and depletion trends. Finally, based on the "learned" well response, a new stochastic method is used for a reliable uncertainty quantification in production forecasting. The proposed workflow has been applied to more than a thousand wells in a major unconventional field successfully through an automated platform, after applying necessary data processing and cleansing procedures. We present value-driven case studies demonstrating tangible ways in which this method can be successfully implemented, providing additional insights compared to traditional forecasting techniques. Our proposed method overcomes common practical challenges of traditional production analysis and forecasting methods (decline curve, analytical and numerical models) to provide a consistent and reliable measure to compare well performance at large scale.