文化基因区间多目标进化优化及其应用

发布时间：2018-04-03 23:04

本文选题：区间参数多目标优化　切入点：进化算法　出处：《中国矿业大学》2017年硕士论文

【摘要】：区间参数多目标优化问题(Interval Multi-objective Optimization Problems,IMOPs)在实际生产、生活中十分常见,而且非常重要,但是由于其目标变量具有不确定参数,使得该类问题很难利用已有的多目标进化优化算法(Multi-objective Optimization Evolutionary Algorithms,MOEAs)对其进行求解。目前,区间参数多目标优化问题已经是进化优化方向的研究重点之一,利用文化基因算法(Memetic Algorithms,MAs)是一种处理该类优化问题的有效方法。此外,利用代理模型去简化局部搜索过程,可以在保证结果精度的前提下,大大提高运算效率。基于此前提,本文给出一种基于代理模型的文化基因区间参数多目标进化优化算法,下面将详细介绍本文算法。首先,本课题在文化基因的算法架构中,融入改进的局部搜索策略,提出一种基于文化基因的区间参数多目标进化优化算法(Interval Multi-objective Memetic Algorithm,IMOMA)。该算法主要包括全局搜索和局部搜索两个部分,全局搜索采用基于区间占优关系的IP-MOEA,局部搜索是本部分的研究重点。研究局部搜索环节时,主要有三个关键技术:局部搜索激活机制,局部搜索初始种群的建立和局部搜索策略。通过10个区间意义下的基准测试函数和太阳能海水淡化中不确定优化问题的计算,并与不含局部搜索策略的IP-MOEA进行比较,得到了收敛性、分布性良好且不确定度小的近似Pareto最优解集,能够说明所提算法IMOMA要优于IP-MOEA。但是由于IMOMA多次使用超体积测度,导致算法时间复杂度过大,运算效率低下。然后,针对IMOMA运算效率低等问题,本文旨在局部搜索中融入代理模型,简化复杂的适应度评价,提出一种基于代理模型的文化基因区间多目标进化优化算法(Surrogate Assisted Interval Multi-Objective Memetic Algorithm,SS-IMOMA)。算法架构依然延续IMOMA的设计,两者的区别主要体现是:在局部搜索策略中,基于个体的超体积贡献与不确定度重新定义个体的适应度函数,并利用支持向量机(Support Vector Machine,SVM)去代理该单目标适应度评价,以达到提高运行效率的目的。同样的是,通过对10个区间意义下的基准测试函数和太阳能海水淡化问题中的不确定优化问题进行优化计算,SS-IMOMA比不含局部搜索的IP-MOEA拥有更好的算法性能,比不含代理模型的IMOMA拥有更小的时间复杂度。最后,利用MATLAB提供的GUI技术设计一个关于太阳能海水淡化问题的回归及优化平台。平台对原始数据的输出进行区间化处理,用来模拟实际工程中的不确定性问题;紧接着,本部分对具有区间参数的数据进行支持向量机回归分析,建立输入与输出之间的映射关系,以此克服实际工程中数值模型难以建立的问题。利用SS-IMOMA可以获得太阳能能海水淡化问题的最优解集,即该问题的最佳运行工况。为了能够方便地修改算法参数,直观地显示运行结果,该GUI平台还提供了算法寻优参数,回归曲线等图像的显示模块,最优解集等表格的输出模块,算法参数设置模块,以及打开文件等控件的设计。综上所述,本文所提的IMOMA能够为区间参数多目标优化问题提供可靠有效的解决途径。针对IMOMA运行效率低的问题,本文所提的SS-IMOMA也成功的解决了此问题。特别是针对区间意义下的基准测试函数与太阳能海水淡化中的不确定优化问题,本文所提的IMOMA和SS-IMOMA都能够得到较好的近似Pareto最优解集。
[Abstract]:The multi-objective optimization problem of interval parameters (Interval Multi-objective Optimization Problems, IMOPs) in the actual production, life is very common, and very important, but because of the target variables with uncertain parameters, so the question is very difficult to use the existing evolutionary multi-objective optimization algorithm (Multi-objective Optimization Evolutionary Algorithms, MOEAs) to solve the problem. At present, the interval parameter multi-objective optimization problem has been one of the key research direction is evolutionary optimization, using genetic algorithm (Memetic Algorithms culture, MAs) is an effective method to deal with the problem. In addition, to simplify the local search process by the agent model, to ensure accuracy, greatly improve the efficiency of operation based on this premise, this paper presents a multi-objective evolutionary optimization algorithm based on cultural gene agent model with interval parameters, The following will detail the algorithm in this paper. Firstly, the algorithm architecture in the cultural gene of this subject, into the local search strategy improved, this paper proposes a multi-objective evolutionary optimization algorithm based on interval parameter gene (Interval Multi-objective Memetic culture Algorithm, IMOMA). The algorithm mainly includes two parts: global search and local search, global search the interval dominance relationship based on IP-MOEA, the local search is the focus of this study. Part of the study of local search links, there are three key points: the local search mechanism of activation of local search, the establishment of the initial population and local search strategies. Through the 10 interval under the benchmark function and uncertainty in solar desalination the calculation of the optimization problem, and compared with and without local search strategy IP-MOEA, convergence is, good distribution and uncertainty in small Like the Pareto optimal solution set, to show that the IMOMA algorithm is better than IP-MOEA. because the IMOMA repeatedly used the hypervolume measure, the algorithm of time complexity, the operation efficiency is low. Then, aiming at the problem of low efficiency of IMOMA algorithm, this paper aims to integrate the local search agent model, simplify the complexity of the fitness evaluation, put forward a multiobjective evolutionary optimization algorithm based on cultural gene region agent model (Surrogate Assisted Interval Multi-Objective Memetic Algorithm, SS-IMOMA). The algorithm is still a continuation of IMOMA architecture design, the main difference between the show is: in the local search strategy, hyper volume contribution of individual uncertainty and redefine the individual fitness function based on, and using support vector machine (Support Vector Machine, SVM) to represent the single objective fitness evaluation, in order to improve the efficiency of the same purpose. Is that through the benchmark functions and the solar desalination problem in the uncertain optimization optimization problem of 10 interval sense, SS-IMOMA has better performance than the IP-MOEA algorithm with local search, has much smaller than that without agent model the time complexity of IMOMA. Finally, the design of a solar seawater desalination problem regression and optimization platform provided by MATLAB GUI technology platform. The original data output interval, used to simulate the practical engineering problem of uncertainty; then, this part of the support vector machine regression analysis of interval parameter data, establishes the mapping relationship between input and output, in order to overcome the numerical model is difficult to establish the problem in practical engineering. The solar desalination can obtain optimal problem solution set can use SS-IMOMA, that is the problem of the operation Condition. In order to be able to easily modify the algorithm parameters, display the operation results, the GUI platform also provides the optimization parameters, such as image display module regression curve, optimal solution set form output module, parameter setting module, design and open the file control. To sum up, the IMOMA can provide reliable and effective solutions for the multi-objective optimization problem of interval parameters. For the low efficiency of IMOMA operation, the SS-IMOMA also successfully solved this problem. Especially for the interval under benchmark test functions and solar desalination of uncertain optimization problems, this paper proposed IMOMA and SS-IMOMA can get better approximate Pareto optimal set.

【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P747;TP18

【参考文献】