A Study Of Multi-objective Particle Swarm Optimization Based On Levy Flight And Its Application

Nowadays,more and more multi-objective optimization problems have appeared in industry and scientific research.Particle Swarm Optimization(PSO)has been applied by a large number of researchers to solve multi-objective optimization problems due to its advantages such as simple implementation,low computational complexity and high efficiency.When solving multi-objective optimization problems,in order to obtain a set of more accurate and well-distributed solutions,many multi-objective particle swarm optimization algorithms(MOPSO)and their variants have been proposed successively.However,once dealing with the multi-objective optimization problems with complicated pareto fronts,some existing multi-objective particle swarm optimization algorithms are easy to fall into local optima,due to their weak global optimization capabilities,resulting in poor performance.Levy flight strategy is a kind of stochastic flight that conforms to levy distribution.It can improve the global optimization capabilities of particles and make them jump out of the local optimal with a greater probability when falling into the local optima.Therefore,the levy flight strategy is introduced into the multi-objective PSO in this thesis to solve the optimization problems in the multi-objective benchmark test functions and the optimization problems in the gene selection.Two improved MOPSO algorithms are proposed for these two different types of problems.The main work of this thesis is as follows:(1)In order to solve the problem that the traditional multi-objective optimization algorithm is easy to fall into the local optima on the complicate multi-objective benchmark test functions,and the diversity of the optimal solution is not good enough,a multi-objective particle swarm optimization algorithm(MOPSO-LFDA)based on levy flight and double archives mechanism is proposed.On one hand,in the iterative process of particles,the levy flight strategy and the particle swarm optimization algorithm are combined to avoid the algorithm falling into local optima.By extending the search range of particles,levy flight can improve the global optimization capabilities of particles and make them jump out of local optima with high probability.On the other hand,when maintaining an external archive,in addition to the primary archive,an additional secondary archive will be created to mitigate the accidental deletion of particles that can result from traditional external archive maintenance methods.Through the proposed double archives mechanism,more useful solutions will be retained,so the diversity of the solution set is increased.In addition,in order to accelerate the convergence rate of the swarm,a novel leader selection strategy is also proposed.In this strategy,solutions that are close to the pareto front and have a large crowding distance are selected as the leaders during the iteration process.Experimental results show that the proposed algorithm is superior to the existing multi-objective optimization algorithms in both convergence and diversity of benchmark functions.(2)In order to solve the problem that the gene selection methods based on traditional MOPSO only focus on the complete search of the non-dominated solutions and is not targeted when searching for the optimal gene subset,which resulting in the poor performance of the selected gene subset on the classifier.At the same time,the selected gene subset still has some redundant genes.This thesis proposes a gene selection method based on multi-objective particle swarm optimization algorithm,which used Levy flight and preference information strategy.First,in order to reduce the dimension of the optimization problem,the method of classification information index is used to filter the original gene expression profile data set,and the number of genes remaining after filtering is used as the dimension of the problem to be optimized to encode the particle,each dimension of the particle represents a gene feature.Then,the classification accuracy and gene size of the selected genes on the extreme learning machine(ELM)are set as fitness functions.Then,different from the leader selection strategy in the algorithm proposed in the previous chapter,the preference information of the decision maker is added to the leader selection strategy,so as to make the selected leader meet the needs of the actual optimization problem better,and then guide the swarm to search in the direction of a better region.At the same time,a variation method based on correlation coefficient is added to the algorithm to reduce the redundancy of the selected gene subset.Finally,the gene subset selected by the proposed PC-MOPSO-LFDA algorithm are put on the classifier of extreme learning machine to verify the classification accuracy of the selected gene subset.Experiments show that the results of the proposed method on five gene expression profile datasets are more competitive than those of several existing gene selection algorithms.