Research On Enhancement Mechanism Of Diversity And Adaptability In Evolutionary Algorithms

With the development of science and technology,the optimization problems in real world are becoming more and more complex.It is difficult to solve them effectively by using traditional optimization methods.Therefore,it is urgent to explore some efficient intelligent optimization methods.Inspired by some phenomena or processes in nature,many evolutionary algorithms(EAs)have been developed and widely used in tackling complex optimization problems.However,as the iteration proceeds,the population diversity of evolutionary algorithms is difficult to maintain,which easily leads to the imbalance between global exploration and local search.In addition,the adaptability of evolutionary algorithms is often poor when solving different types of optimization problems.Motivated by these observations,this dissertation analyzes and explores the diversity and adaptability enhancement techniques of evolutionary algorithms in solving continuous optimization problems from multiple perspectives.Specifically for cuckoo search(CS)algorithm,the adaptive adjustment of control parameters,multi-operator ensemble search and individual guidance mechanism are mainly studied.Specifically,the main research work of this dissertation can be summarized as follows.1)Based on the idea of multi-operator ensemble search and parameter adaptation,a modified CS(MCS)algorithm is proposed.Firstly,an enhancing exploration strategy is employed to expand the search space of each individual.Secondly,according to the percentage that individuals tend to the optimal solution during the last iteration,the step size and discovery probability can be tuned adaptively in the entire search process,which enhances the adaptability of the algorithm to different optimization problems and reduces the probability of falling into local optimum.Moreover,a mutation strategy is employed to further maintain population diversity and alleviate premature convergence,and the global optimal solution information is introduced into biased random walk to improve the local search ability and convergence rate.Finally,to evaluate the effectiveness of the proposed algorithm,the simulation experiments are conducted on 23 benchmark test functions,and several statistical indicators are used for comparative analysis.Numerical results indicate that the proposed MCS has better optimization performance in comparison with five other similar evolutionary algorithms.In addition,to verify the ability of MCS to deal with real world problems,an engineering application example is adopted.Experimental results demonstrate that MCS algorithm is also suitable for solving some practical problems.2)To effectively balance the properties of exploration and exploitation,a novel CS algorithm with multiple update rules,called hybrid CS algorithm(HCS),is developed by using multi-strategy parallel search mechanism.Firstly,to overcome the mutual interference among dimensions and strengthen the local search capability,two one-dimensional update rules with different search characteristics are integrated into CS algorithm framework.Secondly,in view of the occasional long jump characteristics of Levy flight,new candidate solutions can be generated by combining these one-dimensional update rules with Levy flight,which further enhances the exploration performance of the algorithm.Also,a limit value is set in advance to achieve the correct choice among different strategies,which improves the probability of the algorithm jumping out of local optimum.Finally,to fully investigate the convergence performance of the proposed HCS algorithm,49 benchmark test functions with 30 and 50 dimensions are employed,including 11 common problems,10 problems introduced in CEC 2005,and 28 problems presented in CEC 2013.Experimental results indicate that HCS algorithm is better than other CS variants in terms of solution accuracy and robustness,and it also outperforms the seven state-of-the-art evolutionary algorithms.3)No free lunch theorem has proved that no algorithm is suitable for solving all optimization problems.Furthermore,different evolutionary algorithms may have different search characteristics and are suitable for solving different types of optimization problems.For a specific problem,the best generation strategies and parameter settings may vary at different phases of the search process.In response to these challenges,a novel CS algorithm named the ensemble CS variant(ECSV)is developed.Firstly,a candidate pool consisting of three different CS algorithms is constructed.Secondly,according to the previous experiences in producing promising solutions,an adaptive scheme is employed to determine the probability of each CS algorithm being selected.Thirdly,an external archive is embedded to store these discarded solutions in the selection phase,and the random solutions can be selected from the union of the current population and external archive to generate new individuals,which further alleviates premature convergence.Finally,the performance evaluation is performed on 42 benchmark test functions derived from CEC 2005 and CEC 2013.Experimental results show that the proposed algorithm is a competitive method compared with seven CS variants and several other well-known evolutionary algorithms.4)In solving different types of optimization problems,using feedback information reflecting the properties of evolutionary population is one of the important means to enhance the computational efficiency of the algorithm.For this reason,a new CS algorithm with dynamic feedback information(DFCS)is proposed.According to the feedback control principle,the population properties such as fitness value and improvement rate of solution are used as the feedback information to dynamically adjust the algorithm parameters.Firstly,the current population is dynamically divided into three subgroups according to the fitness of individuals,and a self-adaptive parameter adjustment scheme based on cloud model is employed to yield the appropriate step size.Secondly,double evolution strategies are introduced to further balance the tradeoff between exploration and exploitation,and the switching probability between the two strategies is dynamically adjusted according to the improvement rate of solution.Finally,to investigate the efficiency of the proposed algorithm,extensive experiments are conducted on 42 popular benchmark test functions with 30 and 50 decision variables,which are taken from CEC 2005 and CEC 2013.Simulation results indicate that DFCS algorithm has certain advantages in maintaining diversity and enhancing adaptability.5)An improved CS(ICS)algorithm is developed and applied to the vibration fault diagnosis of hydroelectric generating unit.With respect to ICS algorithm,the specific scheme is follows:the biased random walk is modified to make full use of the neighborhood information of the current solution,and a new parameter named scaling factor is introduced;the step size,discovery probability and scaling factor are directly integrated into the search process of the optimal solution,and the appropriate parameter values can be selected automatically in terms of solution quality;the non-uniform mutation operation is used to dynamically adjust the search step of the current optimal solution to further improve the search accuracy;the convergence performance of the proposed algorithm is investigated on 20 benchmark test functions.Then,according to the obtained vibration fault sample data of hydroelectric generating units,a combinational model which ICS combined with BP neural network(ICSBP)is built to identify the vibration faults.Experimental results indicate that ICS algorithm has strong competitiveness in solving these optimization problems.Furthermore,the combinational model can effectively classify the vibration faults of hydroelectric generating units,which enrichs the application fields of CS algorithm and provides a reference for the vibration fault diagnosis of hydroelectric generating unit.