基于Cat映射的多目标猫群优化算法及其应用

发布时间：2018-06-19 12:12

本文选题：猫群优化算法 + 进化多目标优化算法　；参考：《兰州大学》2017年硕士论文

【摘要】：在生活实践和科学研究的实际情况中,许多问题都是具备很大挑战和难度且包含多个优化目标的多目标优化问题(Multi-objective Optimization Problem,MOP)。因为实际需求,多个目标的优化课题吸引了国内外许多研究者的目光,并逐渐变成优化领域的重点钻研课题。进化多目标算法在求解包含多个目标的优化问题时,不但实现简单而且效率较高。本文主要介绍一种比较新的进化算法,即猫群算法(Cat Swarm Optimization,CSO),并对它进行改进并扩展到多目标领域,最后用它优化包含多个目标的问题。CSO算法的理论模型是模仿猫的行为方式创建的,其原理简单易实现,收敛速度较快,算法稳定性较好,已被应用于图像处理、神经网络训练和模式识别等领域并取得了很好的效果。但是由于CSO是近些年提出的,所以理论分析和实践应用方面都需要进行更深入的研究。本文在CSO中引入了精英策略,并将其扩展为多目标,最后用它优化一种简单的改进模型的脉冲耦合神经网络(Pulse-Coupled Neural Network,PCNN)参数实现图像分割。具体的研究工作和创新点如下:1、提出了一种新的进化多目标优化算法:基于Cat映射的非随的多目标猫群优化算法(Non-Random Multi-Objective Cat Swarm Optimization Algorithm Based on Cat Map,NRC-MOCSO)。针对CSO算法在迭代后期极易陷入局部最优和收敛速度缓慢的缺点,本文对CSO做了一点改进,让种群中的猫非随机的进入搜索模式和跟踪模式。再者,针对CSO算法在初始化阶段种群分布不均匀而导致算法不稳定性的弊端,本文利用混沌映射对种群进行初始化,将非随机的CSO与混沌相结合,并将其扩展到多目标领域。2、本文使用提出的NRC-MOCSO算法自动优化一种简单改进模型PCNN(ISPCNN)的参数。首次实现了多目标猫群算法自动优化ISPCNN模型参数。在仿真实验中用以熵为适应度函数的CSO和粒子群优化算法(Particle Swarm Optimization,PSO)、连通性为适应度函数的CSO和PSO、熵与连通性为适应度函数的两个目标的多目标粒子群优化算法(Multi-Objective Particle Swarm Optimization,MOPSO)和NRC-MOCSO六种方法优化ISPCNN参数,用优化后的ISPCNN对五幅经典的图片进行分割实验。结果证明:适应度函数对算法性能有很大的影响;针对于应用问题多目标比单目标更具优势,能够综合考虑多方面的影响因素。
[Abstract]:In the practical situation of life practice and scientific research, many problems are multi-objective optimization problem with great challenge and difficulty and multi-objective optimization problem. Because of the actual demand, the multi-objective optimization project has attracted the attention of many researchers at home and abroad, and has gradually become an important research topic in the field of optimization. Evolutionary multiobjective algorithm is simple and efficient in solving optimization problems with multiple objectives. This paper mainly introduces a new evolutionary algorithm, Cat swarm optimization, and extends it to the multi-objective field. Finally, the theoretical model of CSO algorithm with multiple targets is created by imitating the behavior of cats. Its principle is simple and easy to realize, the convergence speed is faster, the algorithm is stable, and has been applied to image processing. Neural network training and pattern recognition have achieved good results. However, since CSO is proposed in recent years, theoretical analysis and practical applications need to be further studied. In this paper, elite strategy is introduced into CSO and extended to multi-objective. Finally, an improved impulse-coupled neural network named Pulse-Coupled Neural Network (PNN) is used to realize image segmentation. The specific research and innovation points are as follows: 1. A new evolutionary multi-objective optimization algorithm is proposed: Non-Random Multi-Objective Cat Optimization Algorithm based on Cat Map and Non-Random Multi-Objective Cat Optimization algorithm based on NRC-MOCSOO. In view of the shortcomings of CSO algorithm which is easy to fall into local optimum and slow convergence rate in the late iteration this paper makes some improvements to CSO so that the cats in the population can enter the search mode and the tracking mode non-randomly. Furthermore, in order to solve the problem that the population distribution of CSO algorithm is not uniform in initialization stage, the chaotic mapping is used to initialize the population, and the non-random CSO is combined with chaos. In this paper, the proposed NRC-MOCSO algorithm is used to automatically optimize the parameters of a simple improved model PCNNCMOCSO. The automatic optimization of ISPCNN model parameters based on multi-objective cat swarm algorithm is realized for the first time. Multi-Objective Particle Optimization algorithm for CSO and PSO with entropy as fitness function, CSO and PSO with connectivity as fitness function and multi-objective particle swarm optimization algorithm with entropy and connectivity as fitness function in simulation experiments Swarm Optimization / MOPSO) and NRC-MOCSO are used to optimize ISPCNN parameters. Five classic images are segmented by optimized ISPCNN. The results show that the fitness function has a great influence on the performance of the algorithm, and that the multi-objective is more advantageous than the single objective in the application of the problem, so it can comprehensively consider many factors affecting the performance of the algorithm.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】