Research On Evolutionary Algorithm Based On Decomposition Technology To Solve Large-scale Optimization Problems

With the rapid development of technology,many engineering applications contain a large number of decision variables.The optimization of these large-scale problems will challenge existing optimization algorithms.Due to a large number of decision variables,the search spaces of the problems are huge or infinite,which makes it too difficult to solve the problems.Additionally,in the huge exploration space,there must exist many pseudo-global optimal values.These values affect the algorithms and drive it fall into the local optimal value,so the opportunities that to search for the global optimal value would be missed.There are usually two optimization techniques to solve the large-scale optimization.The first method is the cooperative co-evolution optimization framework based on decomposition technology.One problem is divided into several sub-problems firstly in this framework,and the sub-problems are optimized in turn.Finally,the optimization results are combined.The other technique is to address the whole problem directly based on the non-decomposition method.Both optimization techniques are developing rapidly,but according to the "Free lunch" theorem,these algorithms still have some defects.For example,the grouping accuracy is not high,and the performance of the optimizer is not so superior.In order to solve these problems,some new optimization algorithms are proposed.A surrogate-assisted multi-swarm artificial bee colony is proposed.As an optimizer of decomposition technique,this algorithm overcomes the defect of falling into local optimal easily.Firstly,the population is divided into several sub-populations by k-means clustering method,and the information exchange among the sub-populations are adopted after the communication with other sub-populations.In the internal communication,orthogonal prediction strategy is introduced.With the increase of variables,the evaluation times of orthogonal combinations also grow quickly.In order to ensure the utilization of evaluation resources rationally,the known particles are used to evaluate combination locations.The experimental results show that this algorithm performs well.A new cooperative hierarchical particle swarm optimization algorithm is proposed.This algorithm is a complete solution for large-scale optimization.The improved difference grouping method is adopted as the preprocesses to the benchmarks first,and the decision variables in each function are divided into the sub-groups.As the optimizer which containing orthogonal-learning(OL)strategy and level-learning(LL)strategy,the sub-groups are selected into and optimized.CHPSO allocate the computing resources according to the optimization efficiency,and the self-decaying strategy improves the utilization of resources.CHPSO is verified by two large-scale benchmarks,and the experimental results are satisfactory.