Research On Biogeography-based Optimization And Its Application

Evolutionary algorithm is proposed by simulating evolution law in nature, whichrequires knowing few features, such as continuity and differentiability, in solvingoptimization function. Due to its excellent performance for solving functions withnondifferentiability, multimodal, rotated, and noisy, evolutionary algorithm has attractedmuch attention from several scholars.Recently, biogeography-based optimization (abbr. BBO) is proposed based onmimicking migration, mutation of species in biogeography. BBO attracted muchattention for its simple structure and easy implementation. BBO include three operators:migration, mutation, and clear duplicate operators. Migration operator, used forimproving solution quality, is the main operator of BBO, which can take full advantageof information in population. However, migration operator will result in many similarsolutions and decrease exploration ability. Therefore, powerful exploitation ability andpoor exploration ability are main features of BBO, which may decrease the convergencespeed of BBO. To overcome these problems, several algorithms are proposed toenhance exploration ability and accelerate convergence speed. Main research works areas follows.1. Aimed to improve population diversity, an improved BBO based on randomperturbation, abbr. MLBBO, is proposed. First, a hybrid migration operator is designedby integrating improved DE mutation strategy DE/best/2and original migrationequation, which can absorb more informations by both original migration equation andmutation strategy. And then, a random perturbation operator is embedded into BBO forimproving population diversity and enhancing exploration ability. Last,27high-dimension functions are used for numerical simulation. Comparisons between theproposed algorithm and BBO, other improved evolutionary algorithms show thatMLBBO can enhance exploration ability and accelerate convergence speed.2. To improve the exploration ability of BBO, an improved BBO with orthogonallearning, abbr. OXBBO, is proposed. Firstly, an improved migration operator is presentby replacing original migration equation with DE mutation strategy DE/rand/1forimproving population diversity. Secondly, original migration operator mainly searchethe boundary of the search region and neglect the inner of search region, which may bethe promising region. Orthogonal crossover can search both boundary and inner ofsearch region simultaneously. Thence, orthogonal crossover is embedded into BBO for remedying the insufficiency of migration operator and enhancing exploration ability.Thirdly, experimental test are carried out on23functions with different features,experimental results indicate that the performance of OXBBO is better than the others.3. To overcome the problem that much computation and slow convergence speed areresulted by global topology in original migration operator, an improved BBO with ringtopology and Powell’s method, abbr. PRBBO, is present. First, a hybrid migrationoperator is designed by replacing global topology with ring topology and replacingoriginal migration equation with new equation. Second, a new mutation operator is usedto conquer the shortage of the original mutation operator. Third, a self-adaptive Powell’smethod is proposed by using self-adaptive stop criterion and step size in Powell’smethod. And then this method is embedded into BBO for accelerating convergencespeed. Last,24high-dimension functions are used for experimental test, simulationresults show that PRBBO is better than other algorithms in convergence speed andsolution quality.4. BBO with square topology, abbr. ISBBO, is proposed for solving absolute valueequations. As analysis before, a square topology is used to replace global topology forconquering sufficiency of migration operator. A hybrid migration operator is present byintegrating DE/best/2, DE/rand/1, and original equation for acquiring more information.And then, mutation operator in the last proposed algorithm is used for improvingpopulation diversity. Last，eight absolute value equations are solve by ISBBO, theresults show that ISBBO is a effective algorithm.