Research And Application Of Biogeograpy-Based Optimization Based On Self-Organizing Learning

Biogeograpy-Based Optimization(BBO),a recently developed population-based evolutionary algorithm,was inspired by the science of geographical distribution of biological organisms,describing distribution,migration,and extinction of species.BBO has attracted much attention for its unique evolutionary mechanism and excellent searching ability.With the deepening of research and popularization of application,people gradually realize that this algorithm not only has excellent optimization ability,but also exposes the problem of weak local search ability,In view of this,a BBO based on Self-Organizing Map(SOM),denoted as SOM-BBO,was proposed and the problems of complex numerical optimization and chaotic time series prediction are studied in this paper.The main work is as follows:(1)Establishment of self-organizing individual neighborhood learning model.The original BBO uses a global topology in which any two habitats can communicate with each other.If a given habitat is chosen for immigration,any other habitat has a chance to be an emigrating habitat.However,such a migration mechanism makes the location of the optimal solution is easily shared with other individuals,which makes the algorithm is easy to be trapped into the local optima.To overcome this problem,a self-organizing individual neighborhood learning model is proposed.Firstly,the self-organizing network is used to map the population from the high-dimensional space to the low-dimensional space,and the corresponding neurons are found for each individual through repeated training and learning.Secondly,a neighborhood model is constructed for each individual by utilizing the topological invariant nature of the SOM.Finally,the constructed selforganizing neighborhood model is used to achieve the effective transmission of individual information,which is beneficial to maintain the diversity of the population and avoid the algorithm falling into local optimum prematurely.(2)Migration operator design based on self-organizing neighborhood learning.Exploration ability and exploitation ability are two basic abilities of evolutionary algorithm.Exploration ability and exploitation ability are significant indicators of algorightm performance.Efficient evolutionary algorithm is able to search global optimal solution with high precision quickly.Therefore,how to coordinate exploration ability and exploitation ability is the key of algorithm design.Aiming at the strong exploration ability and weak exploitation ability of BBO,a migration operator of self-organizing topology neighborhood learning is proposed.Firstly,according to the constructed self-organizing neighborhood model,a differential migration operator with direction guidance is designed to make the individual evolve in a more favorable direction,which is to improve the local search ability of the algorithm.Secondly,an adaptive selection mechanism is designed to dynamically adjust the computing resource utilization of different migration operators,which is to achieve an effective balance between global and local search capabilities and improve the performance of the algorithm.(3)Research on the effectiveness of self-organizing individual neighborhood learning.In order to verify the effectiveness of the proposed algorithm,this paper selects 4 comparison algorithms to perform simulation experiments on 23 classical test functions,which are commonly used in evolutionary algorithms performance evalution.The experimental results show that overall performance of the proposed SOM-BBO based on self-organizing learning is significantly better than the other ones,and the performance improvement is mainly due to the role of selforganizing individual neighborhood learning mechanism.In addition,the improved algorithm is further applied to the time series prediction problem of chaotic systems.The simulation experiments based on Box-Jenkins and Lorenz chaotic systems demonstrate the effectiveness of the improved algorithm.In summary,the BBO based on self-organizing learning not only significantly improves the performance of the BBO,but also has a certain reference significance for the improvement of other evolutionary algorithms,and has a strong academic value.