骨干粒子群算法及其在电力变压器设计中的应用

发布时间：2018-03-02 14:00

本文选题：工程优化　切入点：粒子群　出处：《浙江大学》2014年博士论文　论文类型：学位论文

【摘要】：随着产业化发展和计算机技术的进步,当前社会进入了大数据时代,出现大量具有维度更高、参数耦合更强、数据量几何级数增长等特征的工程问题。传统的确定性优化算法对此类问题求解有较大局限性,而粒子群(Particle Swarm Optimization, PSO)和骨干粒子群(Bare Bones PSO, BBPSO)算法由于对目标函数要求低、理论与实现简洁、可并行计算、优化性能良好等特征,自提出以来便获得了广泛关注。然而算法以下三点主要问题需要解决：早熟问题广泛存在各种PSO及BBPSO中,同时求解不稳定,算法鲁棒性有待提高；算法并非全局收敛；针对算法求解特性的理论研究过少。本文针对BBPSO的这些问题及其在工程设计中的应用进行了研究,主要内容如下： (一)从最优化问题的求解入手,阐述了PSO的研究背景,详细介绍了BBPSO的发展及算法理论研究、改进研究和应用情况。在对几种典型BBPSO分析的基础上总结了BBPSO的一般形式。 (二)分析了BBPSO算法的客观性和求解特性。针对BBPSO算法的两种不同实现,分别分析了其平移特性、旋转特性和粒子多样性。本文提出并从理论上论证了Ⅰ型实现粒子直线化运动现象和Ⅱ型实现粒子倾向于沿坐标轴运动现象,证明了几种主流分布下的Ⅱ型BBPSO都有坐标轴偏向现象。在这些结论的指导下给出了算法的求解特性以及应用建议。 (三)提出并详细分析了一种基于剪枝策略的骨干粒子群算法(记NPSO)。算法使用了一个基于粒子多样性改进的进化方程,方程中包含一个控制粒子程度的参数。从理论上分析了能确保群体收敛的参数范围并通过实验验证了理论分析；基于随机优化算法收敛性判断准则证明了新方程是全局收敛的；详细描述并从理论上分析了剪枝策略对求解性能的影响,给出了能同时改善算法全局探索能力和局部开发能力的剪枝策略要求。最后标准测试函数实验结果表明算法性能性能相比几种经典PSO算法有显著提升。 (四)研究了基于NPSO算法的电力变压器优化设计问题。详细阐述了变压器优化设计问题,建立了电磁优化设计的数学模型,归纳了四种常用的目标函数以及三类约束条件,并据此分析了问题的解空间。采用NPSO算法实施求解,并提出一种新的约束处理方法以处理多重约束问题。此外设计了一种针对低复杂度变压器设计优化求解的枚举类算法,以该算法和完全枚举法作为对照,指出NPSO算法具有优异的求解性能。 (五)研究了基于NPSO算法的大型电力变压器油箱强度分析。采用有限元方法建立了变压器油箱模型,基于等价弹簧模型模拟上下节油箱连接螺杆在压力下的运动,提升了模型计算精度。以实验数据为基础,用NPSO算法拟合得到弹簧的弹性系数。数值计算结果证实了本文模型的有效性。以该计算模型分析了常用两种油箱类型结构形变和应力,给出最大形变和应力的分布,提出设计应注意要点。
[Abstract]:With the development of industrialization and the development of computer technology, the society has entered the era of big data, with the emergence of a large number of higher dimensions, stronger coupling parameters, engineering characteristics of data geometric growth. The deterministic optimization algorithm for solving this problem has great limitation of traditional particle swarm (Particle Swarm, Optimization. PSO (Bare) and the backbone of the particle swarm Bones PSO, BBPSO) algorithm because of low requirement for the objective function, the theory and implementation of simple, parallel computing, optimization and good performance characteristics, since it is proposed to get wide attention. However, the algorithm the following three main problems to be solved: premature problem widely exists in a variety of PSO and BBPSO at the same time, to solve the instability, the robustness of the algorithm needs to be improved; the algorithm is global convergence; algorithm for characteristics of theoretical research less.
In this paper, the problems of BBPSO and their application in engineering design are studied. The main contents are as follows:
(1) starting from the solution of optimization problems, it expounds the research background of PSO, introduces the development of BBPSO and algorithm theory, and improves the research and application. It summarizes the general form of BBPSO based on several typical BBPSO analysis.
(two) analyzes the objectivity and solution characteristics of BBPSO algorithm. According to the two different implementations of BBPSO algorithm, analyzes their characteristics and characteristics of translation, rotating particle diversity. This paper theoretically demonstrates the type of particle linear movement phenomena and type II particles tend to realize the phenomenon along the axis exercise, that type II BBPSO distribution has several main axis deviation phenomenon. In these conclusions under the guidance of the proposed algorithm and the characteristics of the application suggestion.
(three) this paper proposes a novel backbone particle swarm algorithm based on pruning strategy (NPSO). The algorithm uses a particle equation based on evolutionary diversity improvement, including a parameter control particle degree equation. From the theoretical analysis to ensure that the parameters converge and verified by experiments theoretical analysis; stochastic optimization algorithm convergence criterion proved that the new equation is based on global convergence; detailed description and analysis of the effects of pruning strategies on solving performance theoretically, gives the algorithm can improve the global exploration ability and exploitation ability of the pruning strategies. The experiment results show that the performance of the standard test functions the performance of the algorithm compared with several classical PSO algorithm has significantly improved.
(four) study on the optimum design of power transformer based on NPSO algorithm. This paper expounds the optimization design problem of transformer, establishes a mathematical model of electromagnetic optimization design, summarizes four kinds of objective function and three types of constraint conditions, and analyses the solution space of the problem by using NPSO algorithm. The implementation solution, and put forward a a new method to handle the constraints of multiple constraints. In addition to design a low complexity algorithm for enumeration class transformer design optimization, the algorithm and the complete enumeration method as control, and points out that the NPSO algorithm has excellent performance.
(five) on the analysis of tank strength of large power transformer based on NPSO algorithm. A transformer model by the finite element method, simulation of equivalent spring model on oil tank connecting screw under pressure based on motion, improve the calculation precision of the model. Based on the experimental data, obtained the elasticity of the spring by NPSO algorithm. The numerical results confirm the validity of this model. The calculation model of the two kinds of common types of tank structure deformation and stress are given, the maximum deformation and stress distribution, the design should pay attention to the point.

【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM41;TP18

【参考文献】