Research On Hybrid Swarm Intelligence Optimization Algorithms With Applications

Optimization problems are frequently encountered in the areas of production andlife. Consequently，research on optimization technology with high efficiency has greattheoretical and practical significance. As swarm intelligent algorithms for solvingcomplex optimization problems，these algorithms have received much attention andwidespread applications due to their advantages，such as simplicity，easy to parallelcomputing，intelligent search，etc. In order to further improve the convergenceperformance of these algorithms，it is a really good way to combine the best featuresof different meta-heuristics to derive more powerful hybrid methods. Thus，a fewhybrid swarm intelligent algorithms are developed under the guidance of balancebetween the exploration and exploitation capabilities for unconstrained optimizationproblems including function optimization problems，parameter identification ofchaotic system and data clustering analysis. Generally speaking，the main researchwork of this dissertation is summarized as follows:(1)Aiming at function optimization problems with boundary constraints，first，ahybrid algorithm based on artificial bee colony algorithm and differential evolutionalgorithm，called hABCDE，is proposed. A novel perturbed mechanism and acombinatorial equation of solution search are introduced. Meanwhile，an improveddifferential evolution is also integrated into hybrid algorithm through supplying twokinds of probability about crossover rate. All these enhance the ability of explorationand exploitation of the proposed algorithm. Again，a disaster scheme of population isintroduced to enrich the population diversity. Next，an adaptive multiple strategydifferential evolution algorithm with guiding scheme of Pbest，namely AMSDE，isproposed. In AMSDE，(a) three expert libraries are constructed for differentialmutation strategy，crossover rate together with scale factor according to the researchfindings of previous literatures，(b)for differential mutation strategy with vector Pbest，the vector corresponding to Pbest is randomly chosen one of the top100p%individuals in the current population with a proportion p (0,1]，(c)differentialmutation strategy，crossover rate and scale factor are randomly chosen from theirexpert libraries，respectively，to form three features of the hybrid algorithm. (2)Aiming at parameter identification of chaotic systems，an alternate iterativedifferential evolution algorithm is proposed. Inspired by particle swarm optimization，a novel differential mutation strategy，referred to as DE/pbest/1/bin，is proposed，inwhich the base vector pbest represents the best vector of the target vector in history.Differential mutation strategies，DE/pbest/1/bin and DE/best/1/bin，are used tosimulate the alternate evolving phenomenon of the small evolution process and thebig evolution process in nature. What is more，a new greedy mechanism is employedto further speed up the convergence of the proposed algorithm.(3)Aiming at data clustering problem，a dynamic shuffled differential evolutionalgorithm，called DSDE，is proposed based on a novel technique named randommultiple sampling. In order to better improve the quality of distribution of the initialpopulation，a random multiple sampling method is presented based on weightingmedian value of samples. Differential mutation strategy，DE/best/1/bin，is consideredas the respective differential mutation strategy of two sub-populations to makesub-population evolve independently. Then，a shuffled scheme from shuffled frogleaping algorithm is used as the media of the exchange and sharing of informationbetween two sub-populations to further improve the diversity or convergence speed oftwo sub-populations.A large number of numerical tests and comparisons with some representativealgorithms are conducted in this dissertation to demonstrate the effectiveness of theproposed algorithms.