Many-Objective Optimization Algorithm Applied to History Matching


Abstract

Reservoir model calibration, called history matching in the petroleum industry, is an important task to make more accurate predictions for better reservoir management. Providing an ensemble of good matched reservoir models from history matching is essential to reproduce the observed production data from a field and to forecast reservoir performance. The nature of history matching is multi-objective because there are multiple match criteria or misfit from different production data, wells and regions in the field. In many cases, these criteria are conflicting and can be handled by the multi-objective approach. Moreover, multi-objective provides faster misfit convergence and more robust towards stochastic nature of optimization algorithms. However, reservoir history matching may feature far too many objectives that can be efficiently handled by conventional multi-objective algorithms, such as multi-objective particle swarm optimizer (MOPSO) and non-dominated sorting genetic algorithm II (NSGA II). Under an increasing number of objectives, the performance of multi-objective history matching by these algorithms deteriorates (lower match quality and slower misfit convergence). In this work, we introduce a recently proposed algorithm for many-objective optimization problem, known as reference vector-guided evolutionary algorithm (RVEA), to history matching. We apply the algorithm to history matching a synthetic reservoir model and a real field case study with more than three objectives. The paper demonstrates the superiority of the proposed RVEA to the state of the art multi-objective history matching algorithms, namely MOPSO and NSGA II.