Cooperative Co-evolution Grouping Mechanism For Large Scale Global Optimization

Large scale global optimization problems are a kind of challenging optimization problems with high dimensionality and high nonlinearity.When dealing with large scale global optimization problems,traditional optimization algorithms rely on the mathematical performance of the problem and cannot obtain effective solutions in a reasonable time,while standard evolutionary algorithms can obtain relatively optimal solutions in a short period of time through heuristic strategies.But evolutionary algorithms still cannot solve highdimensional difficulties.Therefore,the Cooperative co-evolution algorithm using the divideand-conquer strategy has become an important algorithm in the field of large scale global optimization recently.The main challenge of using a cooperative co-evolution algorithm is the grouping mechanism,which is an important factor in determining the quality of problem solutions.However,the existing grouping mechanisms cannot efficiently and comprehensively group each part of the problem based on the relationship between variables.In order to solve this problem,this paper proposes a grouping mechanism based on self-adaptive strategy and a grouping mechanism based on interactive contribution.The former grouping mechanism can decompose the different parts of the optimization problem based on the two-phase differential identification method,and then adaptively subdivide the components of different properties effectively.The latter one can cooperate with the covariance matrix adaptive evolution strategy to extract the interactive contribution which is used to group the various parts of the optimization problem.In addition,this paper also proposes different resource allocation methods for the two grouping mechanisms.The self-adaptive resource allocation method is matched with the grouping mechanism based on self-adaptive strategy to improve the optimization efficiency of the components after grouping.The resource allocation method based on contribution distribution is matched with the grouping mechanism based on interactive contribution,which aims to reasonably allocate computing resources to components with different contribution distributions and improve the final optimization results.These resource allocation methods together with the corresponding grouping mechanisms constitute an efficient cooperative co-evolution algorithm.This paper evaluates the effectiveness of grouping mechanisms and resource allocation methods of the proposed cooperative co-evolution algorithm using LSGO benchmarks.Compared with several state-of-the-art algorithms,numerical experiments show that the cooperative co-evolution algorithm in this paper can generate statistically competitive results.Meanwhile,this paper also applies this algorithm to a real world problem.The experiments show that the proposed cooperative co-evolution algorithm is feasible in real world problems and has better experimental performance than other algorithms.