Research On Multi-objective Cuckoo Search Algorithm

Cuckoo search(CS)is an effective population-based swarm intelligent algorithm by simulating the reproductive strategy,and it has been widely applied to solve many engineering problems.However,few of researches focus upon the multi-objective optimization problems(MOP).In this thesis,two efficient cuckoo search algorithms are designed to solve multi-objective problems.For multi-objective problem,a multi-objective cuckoo search algorithm with dynamic local search is designed.In this algorithm,the global search pattern is designed including two non-dominated solutions to provide a reference research direction,while dynamic local search is designed to enhance the local search capability.Furthermore,crowding distance and external population are also employed.To test the performance,ZDT and SCH test suits are employed,and three other multi-objective algorithms are compared,simulation results show our modification achieves the best performance.Many-objective optimization problems are a special case of MOP with objective number larger than three.For this problem,the distance between two adjacent individuals in decision space is very large,and may result in the dynamic changes of fitness.To solve such problem,we propose many-objective cuckoo search optimization algorithm with dimension updating(DUMaOCS).In this algorithm,one probability update manner is designed to limit the fitness changes.Furthermore,non-dominated sorting strategy and reference points strategy are also employed.To test the performance,we compare it with MOEA/D?NSGAIII?KnEA?GrEA and HyPE,simulation results show our modification is superior to other algorithms.The performance of DUMaOCS relies on the Lévy distribution and Gaussian distribution significantly.To verify the influence,we investigate five different distributions: Uniform distribution?Cauchy distribution?Gaussian distribution?Lévy distribution and Exponential distribution.By analyzing the different combinations,we find that the combination of Lévy distribution and Exponential distribution achieves the best performance when compared with MOEA/D? NSGA-III?KnEA?GrEA and HyPE.