Ensemble Method And Its Application Combined With Co-evolution Artificial Fish Swarm Algorithm And SVM

There are always varieties of optimization problems in production and life.Scholars have studied a series of methods on their solutions,among which swarm intelligence optimization algorithm is an important way to solve optimization problems.Artificial fish swarm algorithm(AFSA)is a novel swarm intelligence algorithm,which derived by imitating the foraging,swarming and following behaviors of the fish group.It has a simple theoretical basis,strong robustness,outstanding search performance,easily parallel implementation with other algorithms to improve its performance,and other characteristics.However,AFSA still has some limitations in solving complex problems in practice.Therefore,how to improve the disadvantages of AFSA,at the same time,how to make full use of the advantages of AFSA and integrate with other algorithms to better solve practical problems,which have important research significance.Support vector machine(SVM)has good generalization ability and strong robustness in solving small samples,nonlinearity and pattern recognition,so it has been applied to various fields.However,some hyperparameters of SVM have a great influence on its classification performance,especially the SVM kernel function and its related parameters.Therefore,how to provide a better search strategy for SVM parameter combination optimization,so as to improve the performance of SVM classification and prediction,which has become an important direction of this thesis.Based on the above problems,this thesis mainly studies AFSA and its collaborative ensemble with SVM.The research content is as follows: Co-evolution AFSA(CEAFSA)is proposed by using CEAFSA and SVM.First,an improved AFSA is proposed by initializing evenly distributed population based on good point set theory,and introducing adaptive strategies for visual scope and step and co-evolution strategies among subpopulations.Experiments results on 10 Benchmark test functions indicate that CEAFSA can achieve a good result.Second,the main parameters of SVM are optimized by CEAFSA.Experimental results on 6 UCI datasets demonstrate that the proposed approach has relatively high stability and effectiveness.Finally,haze prediction model is established by using SVM.Experimental results on the haze datasets of Beijing,Shanghai and Guangzhou prove the effectiveness and credibility of the proposed method in the field of haze prediction.