Research On Improvement And Application Of Selfish Herd Optimizer

In recent years,there has been more and more research on nature heuristic optimization algorithm.Compared with traditional optimization algorithms,it has the characteristics of avoiding derivative,simple to implement,fast and effective.This dissertation is based on the selfish herd optimizer,which is a novel nature heuristic optimization algorithm.However,with the advancement of the research on this algorithm,some researchers have found that when solving some realistic optimization problems,the selfish herd optimizer has a small global search range,easy to fall into a local optimal solution,a relatively single population diversity,and weak local search ability.Which leads to the phenomenon of search stagnation,poor convergence accuracy,low quality of candidate solutions and loss of local search ability in the algorithm,and due to the inherent properties of the algorithm,it can't solve discrete optimization problems effectively.The above problems indicate that the algorithm has shortcomings in the structural design,and these shortcomings affect the application scope of the selfish herd optimizer.Based on the above,the main content of this dissertation is to improve the shortcomings of the selfish herd optimizer and apply it to the practical optimization problems.The main purpose is to further improve the optimization performance of the algorithm,improve the structure framework and expand the application scope of the algorithm.The research results are composed of the following five aspects:(1)A selfish herd optimizer fused with Lévy flight search strategy is proposed to overcome the algorithm's search stagnation phenomenon caused by the original algorithm's insufficient global search ability.The Lévy flight search strategy has been proved to be an effective random walk process by many scholars.Each search direction is completely random,and the search step-size also obeys the heavy tailed distribution.At the same time,the search route of this strategy is also widely present in the trajectories of humans and animals,so it is added as a global search strategy to the selfish herd optimizer.The purpose is to expand the search range of the algorithm for the candidate solution space,so as to improve the probability of the algorithm to find a better candidate solution.After that,the improved algorithm is applied to solve the global optimization problem.(2)A selfish herd optimizer fused with simplex search strategy is proposed to overcome the poor convergence accuracy of the original algorithm caused by the low quality of the newly generated candidate solutions.The simplex method is a frequently used and effective optimization strategy in the optimization field.It can be used in the process of heuristic algorithms to generate candidate solutions.In this work,the global optimal solution and the global sub-optimal solution are used as the starting point to perform the search process of the simplex method to generate new candidate solutions,and it replaces the reproduction operation of the original algorithm.So that the quality of the new candidate solution in the improved algorithm is further improved,and the probability of finding better candidate solution in the improved algorithm is increased.After that,the improved algorithm is applied to the data clustering analysis problem.(3)A selfish herd optimizer combining orthogonal design and information update method is proposed to overcome the disadvantages of poor quality of new candidate solutions and single population diversity in the original algorithm.Orthogonal design is a method of selecting representative and high-quality candidate solutions in the candidate solution set,and the purpose of using the information update method is to increase the population diversity of the algorithm,thereby broadening the search range of the algorithm in the candidate solution space,and improve the probability that the algorithm finds the global optimal solution.After that,it is applied to the data set classification problem of the multilayer perceptron.The improved algorithm mainly optimizes the weight value and bias value of the multilayer perceptron,so that the multilayer perceptron can get the best classification accuracy of the data set.(4)A selfish herd optimizer for solving discrete optimization problems is proposed to overcome the defect that the original algorithm cannot solve discrete optimization problems.In this work,a discrete selfish herd optimizer was designed and applied it to solve the graphcoloring problem.Graph-coloring problem is a well-known discrete optimization problem,which mainly includes the optimization problem of "four colors to graph-coloring" and the optimization problem of "minimum number of colors to graph-coloring".In the discrete selfish herd optimizer,a new position update formula is constructed,and the effective strategies to eliminate the conflict area of coloring and the methods to reduce the number of color are added.These strategies and methods effectively improve the performance of the algorithm.(5)A selfish herd optimizer based on piecewise linear chaotic map search strategy is proposed,which overcomes the defect that the original algorithm loses local search ability around the global optimal solution,and makes the improved algorithm find potential better candidate solutions around the global optimal solution.In this work,a piecewise linear chaotic map search strategy is used as a local search mechanism around the global optimal solution.The chaotic sequence of this strategy can obtain many different chaotic search trajectories,and this search trajectory is unpredictable,ergodic,non-periodic and non-repeatability,these characteristics help the improved algorithm to effectively deviate from the local optimal solution.Then,the improved algorithm is applied to IIR filter's identification problem.