Comparative Performance of History Matching Algorithms Using Diverse Parameterization Methods: Application to a Synthetic and North Sea Cases

Abstract

Abstract Extending the life of the Johnston field requires an ability to produce reliable forecasts of the effects of reservoir interventions. Reliable forecasts need an effective, robust and accurate history match. The goal of this paper is to evaluate the relative performance of three different global assisted history-matching algorithms. We describe the implementation of an integrated parameterization and optimization method, which was tested on the Brugge synthetic model (SPE benchmark case study), and the results based on the best selected method are applied to the Johnston gas field in the Southern North Sea. There are two key components to assisted history matching: (i) choice of the parameterization, and (ii) selection and performance of the optimization algorithm. In this paper, three parameterization methods are tested (including use of gradients) on three global history-matching algorithms: Particle Swarm Optimization (PSO), Evolutionary Algorithm (EA) and Differential Evolution (DE). A combination of two different algorithms was also examined. We assessed the algorithm efficiency based on the lowest achieved objective function and the time taken to converge to the lowest value; the quality of the parameterization was examined based on the lowest objective value, the best history match (the most geologically consistent models amongst the lowest objective function models) and the consistency of geological parameters used for the history match. The results show that the effectiveness of each optimization algorithm is dependent on the parameterization method. When comparing parameterization methods, the Around the Median method seems to give the best results in terms of lowest misfit, and best history match when using the same history matching algorithm for the Brugge model example. When comparing across algorithms the DE method performs better than the rest. An iterative combination of algorithms is seen to be the best option and assists in a further minimization of the objective function. The effect of parameterization on match quality is also proved to be important. In some cases, an iterative process can be effective, where the history match is further improved by adding additional parameters or zoning existing parameters. This paper stresses and illustrates the importance of the choice of parameterization methods, while many papers focus only on the performances of the minimization algorithm. It also combines gradient-based parameterization method with global optimization method.