Asaad Abdollahzadeh

A parallel BOA-PSO hybrid algorithm for history matching

In order to make effective decisions regarding the exploitation of oil reservoirs, it is necessary to create and update reservoir models using observations collected over time in a process known as history matching. This is an inverse problem: it requires the optimization of reservoir model parameters so that reservoir simulation produces response data similar to that observed. Since reservoir simulations are computationally expensive, it makes sense to use relatively sophisticated algorithms. This led to the use of the Bayesian Optimization Algorithm (BOA). However, the high performance of a much simpler algorithm — Particle Swarm Optimization (PSO) — led to the development of a BOA-PSO hybrid that outperformed both BOA and PSO on their own.

Multi-objective Scheme of Estimation of Distribution Algorithm for History-Matching

History matching is one of the key challenges of efficient reservoir management. In history matching, evolutionary algorithms are used to explore the global parameter search space for multiple good fitting models. General critiques of these algorithms include high computational demands, as well as low diversity of multiple models. Estimation of distribution algorithms are a class of evolutionary algorithms in which new candidate solutions are obtained by sampling a probability distribution created from the population. In previous works, we studied estimation of distribution algorithms for history matching and showed that good results can been obtained by using a single misfit function. Multiobjective optimisation algorithms use the concepts of dominance and the Pareto front to find a set of optimal trade-offs between the competing objectives of minimising misfit. In this paper, we apply a multiobjective estimation of distribution algorithm to history matching of firstly a well-known synthetic reservoir simulation model and secondly a real North Sea reservoir. We will show that one can achieve higher solution diversity and in some cases better quality solutions by taking multiple objectives. In addition, multiobjective optimisation algorithms are less sensitive to parameter tuning and provide trade-offs between objectives that give more insights into history matching problem.