Hybrid Approach Predicts Accurate Flow Rates and Bottomhole Pressure

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 218762, “Hybrid Virtual Flow and Pressure Metering Approach Integrating Physics and Machine-Learning Models for Accurate and Efficient Rates and Bottomhole-Pressure Prediction,” by Fatemah Al Abudeeb, SPE, Muhammad A. Idris, SPE, and Zulkiflie B. Abdullah, SPE, Saudi Aramco. The paper has not been peer reviewed. _ Accurate and continuous production and pressure measurements are crucial not only for field surveillance but also for effective production optimization. Virtual flowmeters (VFM) are an alternative to conventional physical meters and can exist in physics-based and machine-learning-based forms. The algorithm proposed in the complete paper is a hybrid virtual flow and pressure meter (VFPM) that merges both types. The algorithm’s prediction reduces the time needed to generate data with physics-based VFM while increasing the accuracy of machine-learning-based VFM. Background VFMs involve physics models to predict production flow rates in each well using thermodynamics, fluid dynamics, fluid modeling, and optimization techniques. Physics-based models are often referred to as a white-box models, wherein data is input and transformation of the data is known. The advantages of physics-based VFMs are the need for a moderate amount of historical flow data, a high degree of accuracy, and the ability to run scenarios outside of the history of the well. The disadvantages include the fact that deep knowledge is required to obtain reasonable results and high computational cost. On the other hand, machine-learning-based VFMs, often referred to as black-box models, have the advantage of eliminating previous knowledge, delivering fast predictions, and being scalable. They do not capture well and reservoir evolution over time, however, can result in numbers that do not make sense (e.g., negative flow), and require huge amounts of data to achieve accuracy. The proposed algorithm is the hybrid VFPM, which addresses challenges of flow rates and pressure. The hybrid approach, which the authors refer to as a “gray box,” covers a wide range of operating conditions, even conditions that are not yet seen by the well. Also, contrary to conventional machine-learning-based VFMs, hybrid models build on well-established laws of physics. Overall Modeling Approach The modeling framework begins with basic well data input into the physics-based model (any commercial multiphase-flow simulator) to simulate well behavior and create a sensitivity analysis. Part of the generated data is then fed to the machine-learning model as features. This is because, to predict flow rates and downhole pressure, engineers will have limited data. Although not all data are used as input, the sensitivity analysis for all features are captured; therefore, machine learning will still detect the relationship between features and generalize a model for it. The process then considers correlation between input features and removes highly correlated ones. Data are split into training and testing; then, the regression model is built using multiple model options to choose the most-suitable model based on accuracy metrics and elapsed time. The best model is selected, and target variables are generated.