Some Improved Brain Storm Optimization Algorithms

In 2011,Shi proposed a new swarm intelligence optimization algorithm-brainstorm optimization algorithm.The algorithm used the clustering idea to search for the local optimal,and obtains the global optimal through the comparison of the lo-cal optimal.The variation operation was adopted to increase the diversity and avoid the algorithm falling into the local optimum.In this process,the convergence and di-vergence complemented each other to search for the optimal solution,which was nov-el and suitable for solving the multi-modal high-dimensional functions.At present,it has been used in power,aviation,wireless sensor,financial and other aspects.Howev-er,the traditional brain storm optimization has poor local search ability and its global search ability is also not strong.To improve the performance of BSO on complex problems,a new global version of brain storm optimization algorithm is introduced,which is named as an improved global brain storm optimization(IGBSO).The randomly forward strategy is adopted in the early stage of IGBSO to enhance the swarm diversity,while the all-dimention neighborhood strategy(ADN)is utilized in the later stage.The ADN strategy en-hances the local search capability around the global-best solution to accelerate the convergence speed.Inspired by differential evolution with multi-population based ensemble of mutation strategies,a new variant of BSO algorithm,called brain storm optimization with multi-population based ensemble of creating operations is also proposed.There are three equally sized smaller indicator subpopulations and one much larger reward subpopulation.Three completely different creating operations are used to create indi-viduals.After every certain number of generations,the larger reward subpopulation will be adaptively assigned to the best performing creating operation with more com-putational resources.Finally,in order to improve the efficiency of the MPEBSO,another new algorithm called brain storm optimization with a relaxation selection mechanism is proposed.In the clustering phase,we take random grouping to reduce the computing burden.In the creating ideas phase,a new third creating operation is proposed in this algorithm.In the selection phase,a relaxation selection mechanism instead of greedy selection is proposed.Extensive experiments on the suit of benchmark functions in CEC 2005 and CEC 2015 and comprehensive comparisons with several state-of-art algorithms show the competitive performance of three proposed algorithm.