基于机器学习技术的配电网故障恢复算法研究

发布时间：2018-05-08 02:39

本文选题：机器学习 + 故障恢复　；参考：《湘潭大学》2017年硕士论文

【摘要】：智能电网正在日益成为电网技术的发展趋势,智能配电网属于智能电网当中重要的一部分。智能配电网通常具有较为完备的故障诊断以及自愈功能,以此来提高电网的稳定性以及可靠性,同时智能配电网也支持接入分布式电源。具有优秀自愈能力的配电网能够最大程度减小配电网故障对于用户产生的影响,从而提高用户的用电体验以及配电网的稳定性。配电网的故障恢复是指在配电网的某处发生故障时,通过各个支路的开关以及联络开关进行通断操作从而改变配电网的结构,将受到故障影响的失电区域负载转移到其他馈线或者电源进行供电,从而使失电负载恢复供电。根据配电网故障信息找到最佳的故障恢复路径是故障恢复的主要任务,这是一个典型的多目标非线性问题。本文的主要内容和创新点如下:(1)介绍配电网的构造以及种类,对配电网的故障恢复技术研究现状和存在的技术难点进行了总结和分析,介绍了当前配电网故障恢复的相关理论和一些常用的方案。(2)阐述机器学习理论,以及相关技术在配电网故障恢复上的发展,并提出一种基于回声状态网络(ESN)的配电网故障恢复算法,通过回声状态网络优秀的动态特性对配电网的故障信息进行学习,同时设计树形遍历法结合ESN的输出对配电网的结构进行改变,这种方式能使配电网在更改结构的过程中始终保持辐射状运行条件,因此无需考虑学习过程中由于学习误差而对配电网结构约束造成的影响。(3)在基于回声状态网络的故障恢复方案基础上结合了非支配目标排序遗传算法-2(NSGA-II),通过NSGA-II优秀的全局搜索能力与多目标优化能力使配电网系统进行非监督学习,这种学习方案由于不需要进行训练样本的收集,因此一定程度上减少了时间成本。(4)以标准16节点配电网为实验场景,将本文中研究的基于机器学习技术的故障恢复学习方案应用于配电网的故障恢复学习当中,使配电网系统能够通过学习不同故障状态下的配电网故障恢复方案,从而使系统能够对不同的故障信息进行快速反应。
[Abstract]:Smart grid is increasingly becoming the development trend of grid technology. Smart distribution network is an important part of smart grid. Smart distribution network usually has more complete fault diagnosis and self-healing functions to improve the stability and reliability of the network, and smart distribution network also supports access to distributed generation. The distribution network with excellent self-healing ability can minimize the influence of distribution network fault on users to improve the user's experience of power consumption and the stability of distribution network. The fault recovery of distribution network is to change the structure of distribution network by switching on and off each branch switch and contact switch when there is a fault somewhere in the distribution network. The load of the power loss area affected by the fault is transferred to other feeders or power sources for power supply, so that the power loss load can be restored. The main task of fault recovery is to find the best fault recovery path according to the fault information of distribution network, which is a typical multi-objective nonlinear problem. The main contents and innovations of this paper are as follows: (1) introducing the structure and types of distribution network, summarizing and analyzing the present situation and technical difficulties of fault recovery technology in distribution network. This paper introduces the related theories of distribution network fault recovery and some commonly used schemes. It describes the theory of machine learning and the development of related technology in distribution network fault recovery. A fault recovery algorithm based on echo state network (ESNN) is proposed. The excellent dynamic characteristics of echo state network are used to learn the fault information of distribution network. At the same time, the tree traversal method combined with the output of ESN is designed to change the structure of the distribution network, which can make the distribution network always keep the radiation operation condition in the process of changing the structure. Therefore, it is unnecessary to consider the influence of learning errors on the distribution network structure constraints in the learning process. (3) based on the fault recovery scheme based on echo state network, the non-dominant target ranking genetic algorithm -2n NSGA-IIG is combined, and NSGA-II is used to optimize the performance. The global search ability and multi-objective optimization ability of the show enable the distribution network system to carry out unsupervised learning. Because the training sample collection is not needed, this learning scheme reduces the time cost to some extent.) the standard 16-node distribution network is used as the experimental scenario. In this paper, the machine learning technology based fault recovery learning scheme is applied to the distribution network fault recovery learning, so that the distribution network system can learn the distribution network fault recovery scheme under different fault states. So that the system can respond to different fault information quickly.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM732

【参考文献】