Continuous Bottomhole Pressure Estimation Using Machine Learning and Physics-Based Methods – A Field Case Study

Abstract

Abstract Bottomhole pressure (BHP) has been increasingly integrated into modern workflows in characterizing subsurface reservoir, evaluating well production performance, and optimizing artificial lift designs for unconventional reservoirs. The increasing need for agile asset development planning demands robust and continuous well performance evaluation, for which bottomhole pressure lays the foundation. However, there are several challenges that most of the unconventional operators are facing. It is uneconomical to install permanent downhole pressure gauges and have continuous measurement throughout the entire life span on all wells across the entire asset. A practical approach is to estimate BHP from wellhead pressure by using physics-based multi-phase flow correlations. However, since various multi-phase flow correlations were developed with limited field datasets and assumptions only applicable for certain flow conditions, these empirical or mechanistic models are not generalized to fully characterize the fluid flow behaviors that are applicable for various flow patterns without constant manual selection and tuning. Finally, there is a need for robust estimation of BHP with various changing wellbore configurations under different artificial lift designs and types through the life of the well. In this work, we propose a hybrid methodology integrating physics-based and machine learning models to provide BHP with high accuracy. Five different candidate multi-phase flow correlations were selected for physics-based models to estimate BHP from routine daily production data and consider any change of artificial lift designs and types. With the availability of some downhole pressure gauges to calibrate BHP estimates, we propose to improve BHP estimation in two major steps – first, selecting the best physics correlation for each producing day based on dynamic criteria using a classification method and second, improving the physics-based BHP estimate using a physics-informed machine learning (PIML) approach. The machine learning models were trained based on historical downhole gauge pressure data and validated with data hold-out in history and data measured. The results of model performance showed that this hybrid pre-trained model can be leveraged as a ‘virtual downhole gauge’ to continuously provide high-accuracy BHP estimation in a robust and consistent manner in the absence of physical downhole gauges. In this paper, we present a field case study to demonstrate the deployment and usage of continuous BHP estimation integrating physics-based and machine learning models. This framework has been successfully deployed for one of the largest U.S. unconventional shale basins with over 3000 producing wells. By leveraging this hybrid methodology, high-accuracy and continuous BHP estimates can be provided at field or asset level to streamline well performance analytics workflows for unconventional reservoirs and facilitate better asset development decision-making.