A Study Of Multi-objective Particle Swarm Optimization Based On Quantum Behavior And Its Application

As an efficient swarm intelligence algorithm,particle swarm optimization(PSO)algorithm is widely used to deal with different kinds of optimization problems due to its fast convergence speed and high convergence accuracy.In recent years,a lot of researches and improvements have made PSO achieve good performance when solving single-objective optimization problems(SOPs).However,due to the complexity of multi-objective optimization problems(MOPs),and the weak randomness and intelligence of the traditional PSO algorithm,the performance of the algorithm will be dramatically reduced with the increase of the number of objectives and the dimension of decision variables.One kind of mechanics in quantum world——quantum behavior is beneficial to improve the searching ability of the algorithm.Therefore,quantum behavior is introduced into the multi-objective particle swarm optimization(MOPSO)algorithm,and two improved algorithms will be proposed to solve the multi-objective function optimization problems and gene selection respectively.The main work in this study is as follows:(1)To overcome the deficiency that the classical MOPSO algorithm usually has a bad diversity and convergence when solving MOPs,a multi-objective quantum-behaved particle swarm optimization based on double search strategy and circular transposon mechanism(MOQPSO-DSCT)is proposed in this thesis.Firstly,the double search strategy is proposed to replace the single search pattern in QPSO,which can make the particles mainly learn from their personal best position in earlier iterations and the global best position in later iterations to balance the exploration and exploitation ability of the swarm and improve the diversity.Then,an opposite attractor will be constructed by opposition-based learning when the personal best position of the current particle is equal to the global best position,which can avoid the particle getting into local optimum during local search process;Finally,the circular transposon mechanism will be introduced into the external archive,which can make each dimension of any two particles be exchanged and transferred to generate new particles.This mechanism can greatly enhance the communication ability of the particles in the external archive.The experimental results on eight test functions have demonstrated that this algorithm can produce a set of non-dominated solutions with high accuracy and good distribution.(2)To overcome the deficiency that the classical MOPSO algorithm cannot express decision preference when solving real world problems,a binary MOQPSO-DSCT algorithm based on decision preference(PAG-MOBQPSO-DSCT)is proposed and applied to gene selection.Firstly,IIC method will be used to filter the original data set,the particles will be binary coded according to gene-to-class sensitivity information(GCSI),the classification accuracy and the number of gene subset will be used to design the multi-objective optimization model;Secondly,Hamming distance is introduced to redefine the distance of the particles,which can make the MOQPSO-DSCT algorithm deal with discrete space optimization problems,and the decision preference grid will be introduced into the MOQPSO-DSCT algorithm.One kind of preference fitness function is defined to calculate the fitness of each cube and then the probability of each cube can be calculated according to the preference fitness.Roulette-wheel selection will be applied to select the global best position and maintain the archive,which can make the particles search in the interested region;Finally,extreme learning machine(ELM)will be used as a classifier to verify the classification accuracy of the selected gene subset.The experimental results on five cancer data sets have demonstrated that this method can obtain a gene subset with high prediction ability and low redundancy.