Reservoir model history matching is a multiobjective optimization problem.It involves the adjustment of relevant reservoir model input parameters,to minimize the mismatch between simulated and observed reservoir respo...Reservoir model history matching is a multiobjective optimization problem.It involves the adjustment of relevant reservoir model input parameters,to minimize the mismatch between simulated and observed reservoir responses and to obtain a diverse set of geologically plausible reservoir simulation models.Typically,single objective optimization methods are adopted during history matching.This requires weighted sum scalarization.However,scalarization biases the optimization search,limiting the diversity of the recovered solutions.In this work,a computer assisted history matching procedure based on transform parameterization and a multiobjective evolutionary algorithm with dominance and decomposition(MOEA/DD)is proposed.In the procedure,history matching is treated as a multiobjective optimization problem,parameterized in terms of a small number of kernel principal component analysis(KPCA)variables.KPCA provides efficient parameterization of the reservoir model input property fields.Concurrently,MOEA/DD provides robust and unbiased optimization over multiple objectives.The effectiveness of the proposed procedure is demonstrated with the UNISIM-I-H history matching benchmark problem.展开更多
文摘Reservoir model history matching is a multiobjective optimization problem.It involves the adjustment of relevant reservoir model input parameters,to minimize the mismatch between simulated and observed reservoir responses and to obtain a diverse set of geologically plausible reservoir simulation models.Typically,single objective optimization methods are adopted during history matching.This requires weighted sum scalarization.However,scalarization biases the optimization search,limiting the diversity of the recovered solutions.In this work,a computer assisted history matching procedure based on transform parameterization and a multiobjective evolutionary algorithm with dominance and decomposition(MOEA/DD)is proposed.In the procedure,history matching is treated as a multiobjective optimization problem,parameterized in terms of a small number of kernel principal component analysis(KPCA)variables.KPCA provides efficient parameterization of the reservoir model input property fields.Concurrently,MOEA/DD provides robust and unbiased optimization over multiple objectives.The effectiveness of the proposed procedure is demonstrated with the UNISIM-I-H history matching benchmark problem.
