Data-Driven Based Modelling of Pressure Dynamics in Multiphase Reservoir Model

Abstract

Secondary recovery involves injecting water or gas into reservoirs to maintain or boost the pressure and sustain production levels at viable rates. Accurate tracking of pressure dynamics as reservoirs produce under secondary production is one of the challenging tasks in reservoir modelling. In this paper, a data-driven based technique called Dynamic Mode Learning (DML) that aims to provide an efficient alternative approach for learning and decomposing pressure dynamics in multiphase reservoir model that produces under secondary recovery is proposed. Existing algorithms suffer from complexity and thereby resulting to expensive computational demand. The proposed DML technique is developed in the form of a learning system by first, constructing a simple, fast and efficient learning system that extracts important features from original full-state data and places them in a low-dimensional representation as extracted features. The extracted features are then used to reduce the original high-dimensional data after which dynamic modes are computed on the reduced data. The performance of the proposed DML method is illustrated on pressure field data generated from direct numerical simulations. Experimental results performed on the reference data reveal that the proposed DML method exhibits better and effective performance over standard and compressed dynamic mode decomposition (DMD) mainstream algorithms.