Intelligent Algorithms Design Based On Exploration Ability And Exploitation Ability

Intelligent algorithms are generally inspired by natural phenomena and belong to the stochastic optimization algorithms.Due to their simple structure,ease of use and remarkable global optimization ability,intelligent algorithms have been applied in many domains such as decision optimization,system optimization and industrial design.However,the existing intelligent algorithms still suffer from the problems of premature convergence and stagnation when handling the complex optimization problems.To find out an effective method of alleviating those problems from the viewpoint of arithmetic operating mechanism,based on the balance between global exploration ability and local exploitation ability,this dissertation designs some new improved mechanisms of monkey algorithm or differential evolution,and proposes a novel intelligent algorithm.And the main contributions are summarized as follows.(1)A self-organizing hierarchical monkey algorithm with a time-varying parameter is proposed to improve the performance of the original monkey algorithm.In the new proposed algorithm,the fitness information of individuals,the bounds information of search space,and some new proposed operators(selection operator,fitness-based replacement operator and repulsion operator)are used to redesign the main operations of monkey algorithm,such as climb,watch and somersault.Furthermore,a hierarchical structure is adopted to organize the three main operations,and a self-organizing mechanism is proposed to coordinate them.In addition,a time-varying parameter is employed to replace many fixed parameters of the original monkey algorithm to improve the ease of application.Besides,lots of comparative experiments indicate that the proposed self-organizing hierarchical monkey algorithm performs significantly better than each of the original monkey algorithm and seven intelligent algorithms.(2)A novel variant of differential evolution,which is based on gaussian mutation operator and dynamic parameter,is proposed to enhance the performance of differential evolution(DE)algorithm.In the novel DE variant,the fitness information of random selected individual is applied to design a new gaussian mutation operator and modify a classic mutation operator.Meanwhile,a collaborative rule,which is basedon their cumulative scores,is proposed to coordinate the execution between gaussian mutation operator and modified classic mutation operator.In addition,cosine function and gaussian function are adopted to control the scaling factor periodically and introduce the fluctuation into crossover rate,respectively.Moreover,lots of comparative experiments indicate that the novel DE variant performs significantly better than each of five state-of-the-art DE variants and two other kinds of intelligent algorithms.(3)An improvement mechanism,which is based on re-initialization strategy and adjustment strategy of optimization space,is proposed to enhance the performance of DE algorithm.In the improvement mechanism,a re-initialization strategy is proposed by combining the optimization state of the population and crossover operator to recover the exploration ability.Additionally,to avoid the over exploration produced by the re-initialization strategy,an adjustment strategy of optimization space is designed via utilizing the information of the best individual and a fluctuant parameter.Furthermore,the proposed improvement mechanism is algorithm-independent,and easy to be transplanted into any DE variants.Lots of comparative experiments indicate that the proposed improvement mechanism can significantly improve the performance of a number of DE variants.(4)A novel intelligent algorithm,inspired by the joint operations strategy in military theory,is named as joint operations algorithm and applied to handle large-scale global optimization problems.In the joint operations algorithm,an offensive operation based on the information of elitist and dynamic optimization space is designed to explore the new regions,and a defensive operation based on normal distribution function and crossover operator is proposed to exploit the local regions.A regroup operation based on random permutation is applied to recover the population diversity.Many large-scale global optimization problems are used to perform a comprehensive numerical experiments,and the results show that joint operations algorithm has the best overall performance among the seven compared algorithms.