Self-Learning Reservoir Management

Abstract

Summary In this work, we present an industrial automation framework for control and optimization of hydrocarbon-producing fields while satisfying business and physical constraints. The all-encompassing reservoir-management problem is decomposed into a hierarchy of decision-making problems at different time scales. We exemplify the proposed approach through a case study on a multiple-layer reservoir with a classical waterflood problem, in which a numerical reservoir model is used as a virtual field. A model-predictive control (MPC) strategy is used to regulate well and field instrumentation at economically optimal set points determined by an overlying supervisory control level. The study demonstrates significant reduction in water-handling costs and increased oil recovery. This work is a starting point for further development in automatic intelligent reservoir technologies, which capitalize on the abilities of permanent instrumented wells and remotely activated downhole completions.