Several New Swarm Intelligence Algorithm To Improve Research

The thesis mainly uses swarm intelligence algorithm appeared after2000as researchobject, including artificial bee colony algorithm, shuffled frog leaping algorithm and glowwormswarm algorithm. Based on analysis of algorithm principles, the four algorithms are modifiedfor unconstrained functions optimization or multimodal functions optimization, and simulationexperiments show the effectiveness of improvement strategies.The main contributions of this paper are as follows：1）Artificial Bee Colony (ABC) algorithm has slow speed of convergence in spite of agood global exploration in solving high-dimensional function optimization. In this paper, wemodify the search operator of employed bees and onlooker bees by introducing accelerationcoefficient. Moreover, change the object for employed bees to learn, process roulette strategy ofonlooker bees and admittance of scout bees. Compared with standard ABC algorithm in thesimulation experiment of six benchmark functions with one hundred variables, this algorithmcan get higher convergence precision and cost fewer evolution generations. Experimentalresults show that the improved strategies are effective.2）The core driving power which standard Artificial Bee Colony (ABC) algorithm has insolving high-dimensional function optimization problems is preserved, all kinds of operationsvia which effects are insignificant are canceled, and the algorithm framework is streamlined, sowe propose a simpler ABC algorithm easier to program. Through experimental comparison withsome search operators, we get appropriate one fit for the new algorithm short for SABC. SABCwhich has fewer parameters and faster speed of convergence, can get high precision with smallpopulation size after a few evolution generations. In this paper, six benchmark functions aretested to show good performance of SABC.3）Shuffled frog-leaping algorithm (SFLA) and its passing modified versions are not goodat solving function optimization problems. In order to improve the convergency of SFLA, someideas, which make SFLA a more global exploration, a deeper exploitation, or a smallerpopulation size, are proposed on the basis of its framework and principle. An outstandingversion named F_SFLA, combining these new ideas which prove to be effective, gets excitingsolutions for ten typical benchmark functions. The born boy surprises us, because we can findglobal minima rather than local ones and then work out satisfactory and accurate values relyingon it. Compared with old versions, F_SFLA has faster convergence and better robustness.4）Glowworm swarm optimization algorithm (GSO) always requires large population sizefor solving multimodal function problems to get satisfied solutions. In order to reduce thedampening effect of initial distribution as the breakthrough point, this paper gives fireflies aself-study ability in the initial stage of evolution. And then cut down the cost of exchangebetween individuals by changing operations such as the luciferin update and selection ofmoving direction. After a few representative simulations, as well as comparison with otherliteratures, we can see that this algorithm has great cost advantage in multimodal functionoptimization, because it saves the group size, reduces the evolution iterations, has lessparameters and strong robustness, can capture all peaks and get high precision.