Kernel-based system identification with sparse dictionary learning: applications in petroleum reservoir

Abstract

Closed-loop reservoir management typically requires the use of high-fidelity and com- putationally expensive simulators, where the models need to be executed several times. Proxy modeling consists of a series of methods to build simpler models that aim to reduce computational costs while maintaining adequate levels of accuracy. For the context of reservoir management and optimization, this reduction is crucial and allows the use of techniques that require several simulation iterations. This work proposes a framework for proxy modeling using Kernel-based System Identification and Sparse Dictionary Learning. The models are validated in a synthetic reservoir, with errors be- tween 1% and 2%, and can be used to increase the range of possibilities of control and optimization methods in reservoir management.