【作者】 刘宏达；

【导师】 李殿璞；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 鉴于科学研究和实际工程中许多问题的复杂性、约束性、非线性、多局部极值和建模困难等特点,寻找适用于各种不同需求的新型智能优化方法一直是许多学科的一个重要研究方向。群智能优化技术是模仿自然界群体生物行为特征而产生的一类新兴智能优化算法,该算法在没有集中控制且不提供全局模型的前提下,为求解复杂问题的最优解提供了基础。本文在研究群智能优化算法的模型之——粒子群算法的基本原理和研究现状的基础上,针对基本粒子群算法研究中存在的一些问题,提出了多种改进的粒子群算法,并研究这些改进算法在船舶工程问题中的应用策略。本文的研究目的是探索粒子群算法的改进形式,使之能够有效解决高维大值域多峰函数优化、复杂工业系统的优化控制与决策等理论和工程应用问题。主要研究工作包括以下几个方面:1.针对目前PSO算法参数选取方法(单因素试验法)的不足,提出一种通用的多因素的粒子群参数的选择方法。首先通过粒子群算法参数约束方法来限制参数范围,再利用数论和统计学中的均匀设计法,对算法参数进一步选定。这种试验优选方法,可以在考虑多参数相互耦合、相互制约的复杂条件下,快速给出一组较好的参数值。对参数的泛化能力进行了试验验证。对比分析测试结果表明,本文提出的参数选定方法速度快且优化效果好,对各种改进PSO算法的参数组合选择具有通用性。2.针对高维大值域多峰优化问题寻优困难的现状,提出了基于多种群的空间压缩PSO算法,克服了目前存在的算法在此类问题寻优中存在的缺陷。在PSO算法中引入不完全搜索策略,将搜索分为几个阶段,每一阶段依靠各种群中的最优势粒子来确定各种群下一次搜索的空间。通过对搜索空间的不断压缩,避免了大量无效搜索,提高了搜索的速度和质量。同时,该方法适合多台计算机协同工作,且不需要特殊的并行计算平台。测试表明:本文提出的空间压缩PSO算法可以有效地压缩搜索空间,显著提高搜索效率。3.由于随机算法种群产生的随机性,使得算法的搜索质量和速度也呈随机性,这使普通的随机算法难以满足某些无法多次优化、但又需要实时优化的工程需要。针对这一问题,提出了基于均匀设计法确定关键代次种群的PSO算法。利用均匀设计方法产生PSO算法的初始种群(或关键代次种群),使种群中的粒子在搜索空间分布更好地保持了均匀分散性。给出了4种种群的生成方案,通过测试和对比分析表明,基于值域分割的均匀设计种群生成法能使算法的搜索效果最好;基于均匀设计设定初始种群的算法可以在不丧失搜索质量和效果的前提下,能使算法具有更稳定的搜索效率和搜索质量,同时能有效减少粒子聚集和搜索早熟的随机性发生。4.将社会心理学的意识选择在“个人与团体”、“领导与服从”中的作用引入算法中,给出了一种具有选择意识异步PSO算法。使粒子跟随优势粒子不再盲从而是具有选择能力,并对算法的计算性能进行了对比分析与测试。测试表明,这种结合了社会心理学思想的异步模式PSO算法,可以在保持异步模式收敛速度快于同步模式的优势基础上,进一步提高搜索的质量。5.针对目前船舶自动舵仍信赖PID舵的现状,设计了基于具有选择意识的异步PSO算法的新型实时参数优化。给出了解决该工程问题的步骤和方法和数值计算结果及分析。仿真结果表明:本文提出的PSO免疫自动舵控制策略使得系统鲁棒性得到提高,算法搜索的快速性得到了体现。6.针对目前舰艇消磁目标单一、消磁优化决策难的现状,将基于多种群的空间压缩PSO算法应用到舰船消磁优化决策中,很好地解决了舰船消磁中变量值域范围大、目标函数维数高的工程难题,并可以实现多种目标的优化决策。本文给出了解决该工程问题的方法、步骤和数值计算的结果及分析。本文提出的方法具有通过多阶段的不完全搜索来压缩调节量取值范围、通过多群体之间的独立性保证对压缩区域对最优解包围的概率、通过逐步压缩搜寻范围来减低搜索难度的特点,仿真结果表明,本文提出的方法可以加快消磁优化决策速度和质量,在船舶消磁线圈系统的优化调整工作中起到很好的效果。7.针对船舶电力系统网络重构问题,本文提出了新的离散编码公式,将实数编码的基于均匀设计产生关键代次种群的PSO算法运用到了属于离散优化的船舶电力系统网络重构问题中。不但充分利用了基于均匀设计产生关键代次种群的PSO算法搜索稳定性高的的特点,还形成了一种新的离散粒子群算法。在工程实现过程中,还给出了利用故障提取信息对粒子的取值范围进行压缩的方法和优启发式规则。测试表明:提出的PSO改进算法在决策生成速度上略优于其他方法,搜索质量也和其他方法相当,在最优解的稳定性上明显优于其它算法,对船舶电力系统的网络重构问题有很好的实用性,具有较好的应用前景。最后总结了整个论文研究工作的成果,并展望了粒子群算法需要进一步研究的方向。更多还原

【Abstract】 Researchers have focused on developing novel intelligence optimization methods to address the complexity, constraint, nonlinearity, multiple local minimum as well as the modeling difficulties in lots of scientific and engineering problems. Swarm intelligence optimization technique is such an algorithm developed from mimicking social behavior of animals in the natural environment. The algorithm can be used to solve complicated optimization problem, without requiring centralized control and global modeling.In this dissertation, focus is on particle swarm optimization (PSO), one branch in the swarm optimization. Given the limitations of the existing PSO and its general application fields, several approaches are proposed to revise and improve the existing techniques. And application strategies of these improved algorithms in the field of shipping engineering are investigated.The research in this dissertation is to improve the existing PSO such that it can effectively solve the problems including optimization of complex multimodal problems. The main contributions of the dissertation include:Ⅰ. A kind of method to select PSO parameter based on universal multi-factor is given, which can offset the shortage of normal method to select PSO parameter based on single factor. Basic steps of the method include: in the beginning, parameters bound are limited by the PSO parameter restriction method, then parameters are confirmed by uniform design from number theory and statistics. This kind of method to select optimum parameter can give a group of good parameter quickly in spite of coupling and restriction in the middle of parameters. Otherwise, generalization ability of parameter is researched. Research results show that: the speed of algorithm to select parameter is fast, and effect is good, it is universal to select parameter in all kinds of PSO algorithms improved.Ⅱ. A kind of PSO algorithm based on multi-species space compress strategy is given, which can solve some difficulties in normal PSO algorithm to optimize multidimensional, large range of value and multimodal problems. Incomplete searching strategy is imported to PSO, and searching is divided into several phases. To every phase, next searching space is confirmed by optimum swarm in every a specie. A lot of invalid searching is avoided by compressing constantly space searched, which can improve searching speed and quality. At the same time, the method is suited to multicomputer team working without special team working plat. Test results show that: the algorithm can compress space effetely and improve searching efficiency.Ⅲ. Usually, population is brought randomly in normal PSO algorithm, which leads into randomcity of searching quality and speed. So these normal algorithms can not be used to some engineering problem that can not be optimized many times but real time. In order to solve above problem, a kind of PSO algorithm based on uniform design to confirm key population in some one generation is given. Initial population of PSO algorithm is brought by uniform design, so distributing of swarms in searching space can keep uniform dispersity much better. Four kinds of methods to bring population is given, test results show that: uniform design method based on range of value division can improve searching efficiency.Ⅴ. Consciousness selection in social psychology is imported into PSO algorithm, so a kind of asynchronism PSO with selection consciousness is given, which can make swarm has selection ability not to follow blindly preponderant swarm. Test results show that: compared with synchronization model, the algorithm can keep fast convergence speed and good searching quality.Ⅵ. A kind of immune PID autopilot combining immune algorithm is designed. And the asynchronism PSO algorithm with selection consciousness is used to optimize parameters real time in the new PID autopilot system. Optimization step and method are introduced in a detail. Simulation results show that: the PSO algorithm can improve the robustness of autopilot designed and at the same time, rapidity of the PSO algorithm improved to solve searching problem can proved.Ⅶ. Nowadays, simplex degaussing targets and difficult optimization decision are main shortages in warships and submarines degaussing system. So a kind of new optimization method based on multi-species space compress strategy PSO is applied to optimization decision of warships and submarines degaussing, which can solve engineering difficulties in warships and submarines degaussing such as variable with large range of value and object functions with multimodal. Steps, results and analyses of the method are discussed in a detail. Simulation results show that: the method can improve speed and quality of warships and submarines degaussing optimization decision and bring good efficiency to warships and submarines degaussing winding system optimization.Ⅷ. The electrical network reconfiguration of shipboard power system (SPS) is one of the most important methods to restore power supply and improve survivability. A new kind of discrete coding method based on uniform design PSO algorithm, and the algorithm is used to electrical network reconfiguration of SPS. In real engineering, methods and rules of swarm compress by fault to extract message are given. Test show that: the PSO algorithm improved is better than other methods to improve the speed of building decision, its searching quality is same with other method, but stability of optimum solution is better than others. The method has good practicability and application foreground in electrical network reconfiguration of SPS.更多还原