Research On Social Spider Optimization Algorithm And Its Application

Social Spider optimization(SSO) algorithm is a swarm intelligence algorithm which based on the simulation of the cooperative, communicate through web and mating behavior of social-spider. In the SSO algorithm models,each individual division and cooperation depending on genders. Such fact not only reflects the cooperative behavior of the colony in a realistic way, but also to balance the exploration–exploitation ability of the algorithm to a certain extent.The of algorithm has clear structure, easy to understand and has good search performance. Hence, it has been extensively researched and applied to various fields by scholars at home and abroad. Even so, with the dimensionality of search space increased, the SSO algorithm exists some drawbacks such as local optima entrapment, poor convergence rate and low calculation precision,which had limited it's application.In this paper, aiming to overcome the drawbacks of the social spider algorithm, we improved the SSO algorithm from aspects of the coding scheme and new evolutionary strategy, etc. The improved algorithm was applied to some complex optimization problems. The major purpose is to improve the SSO algorithm performance, perfect its theoretical basis, to broaden its areas of application.The studies in this paper are as follows:(1) A complex-valued social spider optimization algorithm is proposed that improved the SSO algorithm by adopting the idea of complex-valued diploid encoding scheme which expanded the information of social-spider individuals and increased the diversity of the population. Consequently, this strategy can further strengthened the global searching ability of the algorithm, which avoid algorithm trapped in local optimum while improving the accuracy.(2) To overcome the sensitivity to the noise and the initialization, beingtrapped in local optima easily and low performance whiling analyzing the high-dimensional and large datasets, meanwhile the efficiency and effectiveness of the traditional clustering algorithms. We introduced the Simplex Method as a stochastic variant strategy to increase the diversity of the population while enhancing the global searching and local searching abilities of the original algorithm while speeding up the convergence rate and improving the accuracy.The experimental results show that the improved SSO algorithm can improve the accuracy and efficiency of the clustering problem.(3) In view of the traditional mathematical and deterministic methods, it is easy to fall into local optimum, and the population diversity limited when training the feedforward neural network. A novel SSO algorithm with social behivor referenced the PSO algorithm is proposed in this paper. Spider individuals in the improved algorithm have a simple learning ability which greatly enhancing the exploration–exploitation ability of the original algorithm.By comparing and analyzing the experimental results, it is indicated that the improved algorithm is very effective for training feedforward neural network structure.