Improved Differential Evolution Algorithm Based On Population Classification

Global optimization problem widely appears in the economy model,the net-work traffic,image processing and so on.With the development of science and technology,the actual problems are more and more complicated,which enhances the difficulty to solve these problems.Then,it has very important theoretical and application value to research the approach for global optimization problems.Un-like traditional optimization methods that often have many strict conditions in the objective function such as continuity or differentiability,the intelligent opti-mization algorithm can solve the optimization problem by simulating the biolog-ical optimal actions in nature,and has less parameters and fast convergence rate.Consequently,many intelligent optimization algorithms have been proposed to solve the complex optimization problems,such as artificial swarm algorithm,the ant colony algorithm,particle swarm optimization algorithm,differential evolu-tion algorithm and so on.As one of population-based stochastic research algo-rithms,differential evolution owns the benefits of simple rule,strong robustness and less control parameters,and has been widely used in various fields.How-ever,this algorithm still exists the weakness of falling into local optimum easily and slow convergence.To effectively balance the exploration and exploitation,this thesis proposes a differential evolution algorithm based on population clas-sification and its improved version by incorporating the concept of simulated an-nealing,and a generalized tournament selection strategy.In particular,the main content of this thesis describes as follows:1.In order to prevent from falling into local optimum and the slow conver-gence rate,a differential evolution algorithm based on population classification is designed.The proposed algorithm firstly divides the whole population into three sub-populations(superior.general and inferior sub-populations)by means of choosing randomly three individuals from the population and comparing with target individuals according to their fitness values.Then,three mutation opera-tors with different characteristics are assigned for three sub-populations above respectively according to their special individual information,and the control parameters among each mutation operator are suitably adjusted.The proposed algorithm could not only enhance the robustness,but also balance effectively the exploration and exploitation abilities by making full use of the information of individuals.2.To further balance the global exploration and local exploitation ability of DE algorithm,an improved differential evolution algorithm based on popula-tion classification is proposed by introducing the simulated annealing.During the process of dividing population,the target individual needs not only compar-ing with two other individuals selected randomly from the population,but also meeting the corresponding annealing probability.Moreover,an adaptive method is also used to set the control parameters dynamically during the evolutionary process.Therefore,the diversity of population and fast convergence could be improved in the early and late stages of evolutionary process,respectively.3.In order to improve the performance of DE algorithm,a generalized tour-nament selection strategy is developed.First,k(>2)individuals are randomly selected from the parent and its offspring population,and the survival number of the selected individuals n is calculated.Then the top n individuals of them survive into the next generation based on their fitness values;Next,repeat the above process until the population size of the next generation of population is met.Thus,it can improve the exploration and exploitation ability of the algorith-m effectively.Numerical experimental results show that the effectiveness of the proposed algorithms and strategy.