| Cuckoo Search(CS)is a new swarm intelligence algorithm inspired by the behavior of cuckoos in nature.It has been one of research hostspots in the field of swarm intelligence because of its simple structure and easy to implement since it was proposed.Therefore,the CS algorithm has attracted extensive attention of researchers.However,its theoretical research foundation is still not perfect.CS algorithm has still the problem of falling into local optimum or premature convergence.For this reason,this dissertation takes CS algorithm as the main line and learning strategy as the link to improve the algorithm from three aspects: directional learning,kernel learning and extension learning.The main research work of this dissertation can be provided as follows:(1)Firstly,the research status of CS algorithm is described from two aspects of algorithm improvement and application research,including the basic principle of CS algorithm,research status(theory and application),and future research direction and trend.Then,the origin,research background and idea of CS algorithm are described in detail.Finally,the convergence of CS algorithm is analyzed in detail,and the shortcomings of CS algorithm and the problems to be solved are pointed out,which clarifies the direction for the next research in this disseration.(2)Elite opposition learning cuckoo algorithm(CH-EOBCCS)based on chaos perturbation is proposed.The elite opposition learning strategy is introduced to CS algorithm in a dynamic search space.The elite opposition learning strategy is used at a certain probability to generate opposition solution of elite individuals,which form an opposition population to search the corresponding space.At the same time,in order to prevent the algorithm from falling into premature,chaos perturbation strategy is added into the algorithm.The chaos operator is used to perturb the nest position,so as to improve the accuracy of the algorithm.Experimental results show that the performance of CS algorithm can be improved significantly by elite opposition learning and chaotic perturbation.(3)An orthogonal opposition-based learning cuckoo algorithm(TOB-DCS)based on dynamic updating is proposed.In multi-dimensional space,opposition-based learning may lead to individual degradation in some dimensions,away from the optimal position.In this dissertation,orthogonal opposition-based learning strategy is introduced into cuckoo search algorithm,and a certain number of high-quality opposition individuals are generated by orthogonal experimental design,which enhances the exploration of individuals in dimension and effectively improves the diversity of population.At the same time,in order to speed up the local search in the late evolution of the algorithm,a dynamic update strategy is introduced.When a dimension update is completed,it is immediately reconstituted with other dimensions to form a new solution and accept the improved update value.This strategy not only effectively avoids the interference between dimensions,but also improves the local search ability of the algorithm by utilizing one-dimensional useful information.The experimental results show that the TOB-DCS algorithm can significantly improve the quality of the solution and convergence speed.(4)An improved cuckoo algorithm based on adaptive knowledge learning(I-PKL-CS)is proposed.The algorithm extends the traditional single learning optimization model to the multi-learning collaborative optimization model.In the evolutionary process,according to the characteristics of individuals,two learning strategies are alternately implemented,namely,the historical experience knowledge learning and interactive knowledge learning from different sub-populations.The optimal learning mode is selected adaptively.Different learning modes are used to fully explore the different useful knowledge between individuals and sub-populations.These knowledges are used in the subsequent evolutionary process to guide the later search of the algorithm.It effectively improves the search efficiency of the CS algorithm.The proposed algorithm is verified on 28 benchmark functions and typical engineering design problems,respectively.The experimental results show that the proposed I-PKL-CS algorithm has better balance between exploration and exploitation on benchmark function and practical engineering problems.(5)Another improved cuckoo search algorithm based on Q learning(MP-QL-CS algorithm)is proposed.Parameter settings have great influence on the performance of the algorithm.A single parameter control strategy will make the individual lack of extensive adaptability.Fixed parameter setting method will not be effective in every step of evolutionary iteration all the time.In this disseration,a multi-step Q learning strategy is introduced.At each iteration,different parameter strategies are used to learn forward.Individuals choose the optimal parameter strategy according to the learning effect for the next evolution.At the same time,in order to avoid falling into local optima and oscillation near the local optimal position,the population is divided into two subpopulations,and Q learning and adaptive strategy are used to co-evolve,which effectively improves the accuracy and convergence of the algorithm.The simulation results show that MP-QL-CS algorithm has excellent performance on single-modal function,multi-modal function,and complicated rotated function.In summary,this disseration takes the model as the starting point,theoretical analysis as the focus,typical problems and engineering application problems to solve as the foothold,forming a three-point-one-line algorithm design and analysis method,and proposes a learning-based cuckoo algorithm model.Through combining domain knowledge,the dissertation is to design efficient algorithms,which not only enriches the theoretical research of cuckoo algorithm,but also broadens the related application fields of cuckoo algorithm.At the same time,it improves the application value of swarm intelligence algorithms,and provides new research ideas for the development of the theory and application research of evolutionary computing and natural computing. |
