Improvement And Application Of Marine Predator Algorithm

The Marine Predator Algorithm(MPA)is inspired by foraging strategy of marine organisms based on Lévy and Brownian movements,as well as the optimal encounter rate strategy between predator and prey.MPA has the advantages of simple structure,few parameters and easy implementation of program,etc.However,the algorithm still has some deficiencies such as easy to fall into local optimal,slow convergence speed and low optimization accuracy,etc.This thesis analyzes and overcomes the defects of MPA,and applies the improved algorithm to feature selection,air quality index prediction and cloud task scheduling.The main work of this thesis is divided into three aspects:(1)The feature selection method of rough set model can’t directly handle continuous data,and MPA still has some problems,such as slow convergence speed and easy to fall into local optimal,etc.Therefore,a feature selection method based on MPA and neighborhood rough set(NRS)is proposed.Firstly,the initial solution based on Tent chaotic map and opposition-based learning is adopted to improve the diversity of the initial solution.Secondly,in order to overcome the slow convergence speed and jump out of local optimization of the algorithm,Gaussian perturbation strategy is integrated into MPA.Thus,an improved marine predator algorithm(IMPA)is proposed.Then,a new feature selection algorithm is developed by combining IMPA and NRS.Finally,the experimental results of UCI datasets show that the proposed feature selection algorithm has significant performance.(2)Aiming at the sensitive issue of parameter selection in Extreme Learning Machine(ELM)model and the shortcomings of MPA,an enhanced marine predator algorithm(EMPA)is proposed to optimize the parameters of ELM.Firstly,a quasi-reflexive learning strategy is introduced to generate high-quality initial solution.Secondly,the Cauchy mutation strategy is introduced to update the search agents to enhance the optimization ability of the algorithm.Then,the crossover strategy is introduced to modify the search agents to further improve the optimization accuracy of the algorithm.Finally,EMPA and ELM are combined to predict the Air Quality Index(AQI).The experimental results show that the proposed algorithm is superior to the comparison algorithm.(3)In order to expand the application of MPA,a hybrid Marine Predator Algorithm(HMPA)is proposed.Firstly,the operator of Whale Optimization Algorithm(WOA)is introduced into MPA to improve the exploration capability.Secondly,nonlinear inertia weight coefficient is integrated into MPA to a better balance of exploration and exploitation.Thirdly,golden sine strategy is utilized to improve the convergence accuracy and speed of MPA.Finally,HMPA is applied to the cloud task scheduling problem.The experimental results show that the proposed algorithm is superior to the comparison algorithm in terms of load balancing and resource utilization.