Study Of Meta-heuristic Algorithm For Large Scale Global Optimization Problems

In real life,many engineering problems,numerical problems and combinatorial problems are abstracted as global optimization problems to solve.However,some global optimization problems have hundreds of decision variables.Due to the existence of "Curse of dimensionality",some classical optimization algorithms are not able to solve these problems properly.Therefore,many scholars have listed this kind of optimization problem separately,which is called large-scale global optimization problem.As a general solution of optimization problem,heuristic algorithm has the characteristics of fast convergence speed and strong global search ability.Good results are obtained in the low dimensional optimization problem.However,the performance of these classical heuristic algorithms on the large scale global optimization problem is decreasing rapidly.Cooperative coevolution cooperative coevolution framework is a popular solution for large scale global optimization problem.The core idea is to decompose the large-scale optimization problem into several sub optimal problems,and in the sub optimization problem with heuristic algorithm better performance optimization.With respect to this divide and conquer strategy,some scholars through the heuristic algorithm,combining the classic local search strategy is added for overall research on large-scale global optimization problems,and achieved a series of research results.In this paper,based on the two typical research methods of large scale global optimization problem,an in-depth study is carried out.(1)An opposition-based learning Competitive Swarm optimization(OBL-CSO)algorithm is designed.This algorithm can effectively extract the new search area in later iterations to increase the probability of getting out of the local optimum;the numerical results show that the OBL-CSO algorithm has faster convergence speed of global optimization problems in small and medium scale global optimization problems,higher convergence precision and efficiency in large-scale global optimization problems significantly.(2)A Frequently Coverage(FC)strategy is designed.This algorithm produces a new searching subspace in each iteration,then use RDPSO algorithm to get Pbest of each particle in the subspace.Finally,all the particles update their position in the subspace to their Pbest.Through iteration,all searching subspaces intersect with each other frequently,covering the entire searching space,and then get the global optimal value.Combined with the Random Drift Particle Swarm Optimization(RDPSO)algorithm and Particle Swarm Optimization with quantum behavior(QPSO).We propose two algorithms: FC-RDPSO algorithm FC-QPSO algorithm finally.The algorithm is tested on several well-known benchmark suites,and a large number of numerical experiments show that the FC-RDPSO and FC-QPSO algorithm not only has good effect,but also has high efficiency.(3)Self-adaption Differential Grouping(SDG)algorithm is designed,which is a new automatic grouping method of decision variables.This method greatly improves the accuracy of grouping,and the accuracy rate is 100% on the CEC'2010LSGO benchmark suite;The optimization stage,we design the Pyramid Allocation(PA)strategy and the Self-adaptive Pyramid Allocation(SPA)strategy to allocate cost of fitness evaluation.The CCPA-SDG algorithm and CCSPA-SDG algorithm are combined by grouping algorithm and fitness evaluation cost allocation algorithm,simulation experiments show that the CCPA-SDG algorithm and CCSPA-SDG algorithm has a strong ability to optimize the large-scale global optimization problem has a very good effect.In this paper,we study the Competitive Swarm optimization algorithm and cooperative coevolution framework.The strategy of frequent coverage and adaptive grouping is proposed.The two problem decomposition methods are applied to the classical heuristic algorithm,and a series of algorithms based on cooperative coevolution are also proposed.Has a good performance in the 1000 dimension large-scale global optimization problems,in addition to the combination of opposition-based learning mechanism presents an opposition-based learning Competitive Swarm optimization algorithm,this algorithm not only in the low dimensional optimization problem on the optimization performance significantly,but still maintain its optimization performance in large-scale optimization problems.