关键词: L-M算法, 差分进化算法, 密度聚类, 混合方法, 试井参数反演

Abstract: Parameters optimization of well test optimizes the reservoir parameters that are inversed by the information，which is based on the varieties of measured bottom pressure or flow with the change of time.The complex equations and boundary conditions make the problem of well test parameters optimization be high nonlinearity and existence of many local extremums in modern well test.This paper presents a kind of hybrid method which is based on L-M algorithm and differential evolution algorithm.Hybrid method uses differential evolution algorithm to get characteristics of population clustering after certain evolutionary generations and identifies the populations as different clustering regions.And then by making the center of each clustering as the starting point，it re-uses L-M algorithm that is based on the gradient of local search with strong ability of local search to quickly find the least value in this clustering region.Hybrid method combines the advantages of both differential evolution algorithm with strong global search ability and L-M algorithm with strong local search ability as well as high speed of convergence.The paper applies this hybrid method into well test parameters optimization and demonstrates that it has higher optimizing speed and exacter convergence precision than other any single algorithm through providing the results of two different kinds of oil reservoir model.Besides，the hybrid method has wide practicability and can efficiently solve the complex issue about the existence of many local extremums.

Key words: L-M algorithm, differential evolution algorithm, density clustering, hybrid method, parameters optimization of well test