基于群集智能的复杂问题优化算法与应用研究

发布时间：2018-04-06 20:17

本文选题：群集智能　切入点：大规模优化　出处：《武汉大学》2016年博士论文

【摘要】：“创新、协调、绿色、开放、共享”是“十三五”时期乃至中长期指导我国能源电力行业科学发展的新理念。在着力推进能源电力行业创新发展与绿色发展的进程中,大量亟待优化与创新的技术问题相继涌现,且随着电力系统规模的日益增长、技术要求的不断提升,这类技术问题呈现出规模化、复杂化的发展趋势。本文依托实际科研课题,以群集智能思想的应用为出发点,结合专业背景,围绕电力系统建设、电力系统运营中的两类典型复杂优化问题展开研究：大规模光伏系统复杂光照下最大功率点跟踪以及电能计量设备运维作业动态优化,抽象出一类具有大规模、多极值、变量耦合等特性的复杂优化问题,并建立基于群集智能的求解模型。在此基础上,针对不同问题的属性与特点研究基于群集智能的求解方法,并最终回归实际问题的求解与优化。具体地讲,本文主要研究内容及创新成果如下：对于具有多极值特性的复杂优化问题,由于群集智能算法易出现因个体陷入局部极值且难以摆脱而导致的“早熟”收敛现象,极大程度地限制了算法对于这类问题的求解性能。本文以粒子群算法为例,分析其“早熟”现象的形成原因,并从增强粒子个体智能属性的角度出发提出若干防“早熟”策略以及HSPSO、 HSPSO-FI算法,通过为个体引入仿人脑的智能属性以增强其摆脱局部极值点束缚的能力。仿真实验表明,通过引入仿人智能属性,粒子个体能够有效克服局部极值点的束缚,算法优化性能得以显著提升。对于具有大规模特性的优化问题,由于问题复杂度随变量维数的增加呈指数上涨,这一“维数灾难”的出现将导致常规优化算法失效。尤其当大规模优化问题同时具有变量耦合特性时,问题的求解将变得极为复杂。为拓展群集智能的应用领域,提升其对各类大规模优化问题的求解性能,本文研究并提出一类通用的多参考向量自适应协同进化(AM-CC)算法框架,并以粒子群算法为例提出AM-CCPSO算法。仿真实验表明,AM-CC框架对于具有变量可分割以及变量不可分割等特性的1000维大规模问题具有良好的求解性能。AM-CC框架的提出为群集智能应用于求解大规模问题,尤其对于具有变量耦合特性的大规模问题求解提供了一种通用、有效的解决方案。在上述理论研究的基础上,针对电力系统建设中的典型复杂优化问题展开应用研究：围绕大规模光伏系统复杂光照下的“热斑效应”与最大功率点跟踪问题,研究并提出了基于群集智能的求解方案。“热斑效应”对光伏系统局部遮阴环境下的稳定工作构成严重威胁,现有方法普遍存在系统输出功率额外损失、成本较高或难以在大规模系统中应用等缺陷。针对这一问题,本文研究了基于光伏电池控制装置与支路稳压装置的大规模光伏阵列拓扑结构,为实现单块电池板(或最小控制单元)级的最大功率点跟踪提供了硬件基础。此外,建立了以大规模优化问题为描述形式的大规模光伏系统最大功率点跟踪数学模型,并将本文理论研究部分提出的各算法应用于模型求解。仿真实验表明,通过拓扑结构、数学模型与求解算法的相互配合,大规模光伏系统各电池板(或最小控制单元)在复杂光照环境下能够稳定工作于各自理论最大功率点,使“热斑效应”得以有效解决的同时保证了系统的最大输出功率。此外,针对电力系统运营中的典型复杂优化问题展开应用研究：围绕电能计量设备运维作业动态优化问题,分析电网企业相关管理工作的实际需求,并建立基于群集智能的运维作业动态优化模型,以实现对任务点数量、实时路况、运维人员属性与数量、决策者偏好等外部条件的实时响应。在此基础上,采用本文理论研究部分提出的各算法完成对模型的求解。仿真实验表明,提出的模型与算法能够对电网企业关于运维作业的各项要求予以实时响应,实现电能计量设备运维作业的高维度实时、动态优化,提升电网企业日常运维工作管理效率。
[Abstract]:"Innovation, harmony, green, open, sharing" is a new concept of "13th Five-Year" period and long-term guidance of scientific development of China's energy and power industry. In order to promote the development of innovation and development of green energy power industry in the process, many technical problems need to be optimized and innovation have emerged, and with the increasing scale of power system the technical requirements of the continuous upgrading, the technical problems showing a large-scale, complex trend. Based on the practical research project, by using the swarm intelligence theory as the starting point, combined with professional background, focus on the construction of the power system, power system operation in the two typical complex optimization problem is studied: mass photovoltaic system under complex illumination maximum power point tracking and metering equipment operation and maintenance of dynamic optimization, abstract a class has a large-scale, multi peak, variable coupling etc. The complex optimization problems, and establish a model based on swarm intelligence. On this basis, according to the characteristics of different problem solving method based on swarm intelligence, solving and optimization and finally returns to the practical problems. Specifically, the main contents and innovations are as follows: for the complex optimization problem with multi extremum the characteristics, due to swarm intelligence algorithm is prone to fall into local extremum and because the individual is difficult to get rid of the "premature convergence" phenomenon, greatly limits the performance of algorithms for solving this kind of problems. This paper uses the particle swarm algorithm as an example, the cause of formation of the "premature" phenomenon, and starting from the enhanced particle individual intelligence attribute aspect puts forward some anti "premature" strategy and HSPSO, HSPSO-FI algorithm, through the intelligent properties into humanoid brain for the individual to enhance their escape from local The ability of extremum bound. Simulation results show that through the introduction of intelligent property, the individual particles can effectively overcome the constraints of local extremum, algorithm optimization performance can be significantly improved. For the optimization problem with large scale property, due to the complexity of the increase with the dimension of the exponential rise, the "curse of dimensionality" will lead to the failure of the conventional optimization algorithm. Especially when the large-scale optimization problems with variable coupling characteristics, solving the problem will become extremely complex. For the expansion of the application of swarm intelligence, to enhance its various types of large-scale optimization problem solving performance, a generic multi reference vector adaptive co evolution is studied in this paper and put forward (AM-CC) the algorithm framework, and particle swarm algorithm as an example. Simulation results show that the proposed AM-CCPSO algorithm, the AM-CC framework for variable segmentation and variable cannot be divided The 1000 dimensional cutting characteristics of large-scale problem solving performance has good.AM-CC framework for the application of swarm intelligence to solve large-scale problems, especially provides a general for solving large-scale problems with variable coupling characteristic, effective solution. Based on the above theoretical research, applied research on typical complex optimization problems in power system construction: focusing on the large-scale photovoltaic system under complex illumination "hot spot effect" and the maximum power point tracking problem, study and put forward the solution scheme based on swarm intelligence. "Hot spot effect" poses a serious threat to the stability of the photovoltaic system partial shade environment, the existing methods are additional system output power loss, high cost or difficult in large scale applications and other defects. To solve this problem, this paper studies the control based on photovoltaic cell Large scale photovoltaic array device topology and branch voltage stabilizing device, to achieve single panels (or minimum control unit) provides hardware based maximum power point tracking level. In addition, to establish a large-scale optimization problem described in the form of large-scale photovoltaic system maximum power point tracking model, and some theoretical research of this paper is put forward the algorithm is applied to solve the model. Simulation results show that the topological structure, interaction mathematical model and solution algorithm, the system of large scale photovoltaic panels (or minimum control unit) in complex illumination can work stably in the respective theoretical maximum power point under the environment, the "hot spot effect" can also effectively solve the maximum output power of the system. In addition, the application of typical complex optimization problems in power system operation: around the power metering equipment. The dynamic optimization problem of dimension operation, analysis of the actual needs of the relevant management of power grid enterprises, and set up the model of dynamic optimization of operation and maintenance based on swarm intelligence, the number, in order to achieve the task of real-time traffic, the number of attributes and operation and maintenance personnel, real-time response preference and other external conditions. On this basis, using the algorithm in the part of theory to study and put forward the solution of the model. Simulation results show that the proposed model and algorithm can be real-time response to the power grid enterprises on the operation and maintenance requirements, implementation of electric energy metering equipment operation and maintenance of the high dimension of real-time, dynamic optimization, enhance the daily maintenance work efficiency of management of power grid enterprises.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP18

【相似文献】