CASIA OpenIR  > 毕业生  > 博士学位论文
信息驱动的电网静态电压稳定态势评估方法研究
白熹微
Subtype博士
Thesis Advisor谭杰
2020-06-02
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Name工学博士
Degree Discipline控制理论与控制工程
Keyword复杂电网 静态电压稳定 态势评估 信息驱动 信息融合
Abstract

电力网络的安全稳定运行事关国计民生。随着社会用电需求的日益提升,现代电网的覆盖范围和复杂程度持续增长,在系统宏观层级和元件微观层级上分别呈现出高维非线性和耦合随机性的特点,导致建模仿真的困难度提高,电网出现电压失稳事故的风险逐步升级。传统基于机理/模型的静态电压稳定分析方法在准确性和时效性上存在较多问题,面临迫切转型。一方面,电压失稳关键环节得不到准确定位和提前干预,导致电网防控操作的盲目性和被动性;另一方面,电网电压整体稳定态势复杂多变,态势评估的滞后性延误了防控决策的关键时间,导致稳定态势恶化甚至引发大停电事故,造成严重的社会和经济损失。本论文针对上述缺陷,围绕复杂电网静态电压稳定态势评估问题,从信息驱动的角度出发,深度挖掘电网海量实测数据中的关键信息,融合机理模型与运行经验中蕴含的可行规律,研究电网静态电压稳定状态判定、裕度估计和关键环节识别方法,从而提升电网的防控能力并降低电压失稳和崩溃风险,有效预防大停电事故。论文的主要工作和创新成果归纳如下:
(1)提出了一种考虑网络拓扑信息的电压稳定裕度估计方法,基于节点相似性设计了蕴含拓扑信息的局部结构,将稳定裕度估计类比于图像理解,通过图嵌入技术将非结构化的拓扑信息转化为局部结构中的结构化向量,融入节点运行状态信息作为深度卷积神经网络的输入,进而逐级融合“局部”和“全局”特征形成高阶表达以构建基于实测信息的稳定裕度实时估计模型。实验结果表明,局部结构的引入能够在多个层面有效降低估计误差,从而证实了所提方法的优异性能。本研究面向单一正常态或故障态网络,从定量分析的角度实现了“模型+信息”驱动的静态电压稳定态势评估。
(2)提出了一种基于区域信息集成的电压稳定状态判定方法,针对线路故障导致网络拓扑结构变化进而影响稳定状态的问题和因故障传播特性导致相同稳定状态存在多种负荷分布亚型的问题,根据源节点能量传输路径的相关性划分电网子区域并引入介数中心性指标对网络拓扑变化程度进行量化表达,通过集成子区域状态信息、拓扑变化信息和故障描述信息的多层级深度神经网络实现了基于实测信息的电压稳定状态多级别分类。实验结果表明,所提方法对信息构成和模型结构的优化能够明显提升判定精度。本研究面向多故障态网络,从定性分析的角度实现了“模型+信息”驱动的静态电压稳定态势评估。
(3)提出了一种经验信息驱动的无监督电压稳定状态评估方法,针对电网系统和元件级的复杂性与随机性提升导致建模困难的问题,设计了经验规则引导数据降维聚类的新方法,基于孪生自编码器构建用于评估电压稳定状态的最优潜空间,进一步结合聚类算法实现了不依赖机理模型的电压稳定等级划分。实验结果表明,所提方法能够有效逼近理论计算的真值。本研究基于无监督学习,探索实现了信息驱动的无模型定性静态电压稳定态势评估模式。
(4)提出了一种融合时空信息的电网关键节点和区域识别方法,采用节点电压相轨迹的变化程度作为节点重要度和关联度的衡量指标,无需电网元件模型和拓扑结构参数,从时间和空间两个角度提取电压相轨迹的运动学和形态学属性信息,基于两阶段信息融合与聚类分析方法动态识别影响电压稳定态势的电网关键节点和区域。实验结果表明,所提方法的识别结果能够准确反映电压稳定态势的变化。本研究从溯源分析的角度出发,实现了信息驱动的静态电压稳定关键环节搜索和定位,形成整体研究的闭环。
(5)集成以上主要关键技术,设计并实现了智能电网在线电压稳定态势评估原型验证系统,阐述了其总体架构与核心功能并对本论文所提方法的有效性进行了应用验证。

Other Abstract

The security and stability of power grid is highly related to the national economy and people's livelihood. With the increasing social demand for electricity, the coverage and complexity of modern power grid increase continuously, characteristics of high-dimensional nonlinearity and coupling randomness are respectively presented at the macro level and the micro level of the system and the components, resulting in the increasing difficulty of modeling and simulation analysis as well as the upgrading risk of voltage instability accident in the power grid. Traditional mechanism/model-based power grid static stability analysis approaches are flawed in their poor accuracy and real-time performance, thus facing urgent transformation. On the one hand, the critical elements of voltage instability cannot be accurately located and intervened in advance, which leads to blindness and passivity of the power grid prevention and control operation; on the other hand, due to the complex and changeable overall voltage stability situation of power grid, the hysteresis of situational assessment delays the prime time of prevention and control decision-making, leading to the deterioration of stability situation or even large-scale blackouts, which may cause serious social and economic losses.In view of the above defects, this dissertation focuses on the problem of static voltage stability situational assessment of complex power grid. From the perspective of information driven theory, the key information in the massive measured data is deeply excavated, the feasible laws contained in the mechanism model and operation experience are integrated. We studies the approaches of power grid static voltage stability state determination, margin estimation and critical element identification so as to improve the prevention and control ability of power grid, reduce the risk of voltage instability and collapse and prevent large-scale blackout effectively. The main work and innovation achievements are summarized as follows:
(1)An approach of voltage stability margin estimation with consideration of the grid topology information is proposed. Based on the similarity among nodes, a local structure containing topology information is designed. The estimation of voltage stability margin is compared with image understanding. The unstructured topology information is transformed into structured vector in local structure by graph embedding technology, and the operational information of node is integrated as the input of deep convolution neural network. Then, the "local" and "global" features are fused hierarchically to form a high-order representation for building the real-time voltage stability margin estimation model based on measured information. Experimental results show that the introduction of local structure can effectively reduce the estimation error at multiple aspects, which proves the excellent performance of the proposed approach. Facing a single normal or fault state, this study realizes the static voltage stability situational assessment driven by the "model+information" from the perspective of quantitative analysis.
(2)An approach of voltage stability state determination based on regional information integration is proposed. Aiming at the problem that the network topology changes caused by line fault affect the stability state as well as the stability state has multiple load distribution subtypes due to the fault propagation characteristics, the proposed approach divides the network into sub-regions according to the correlation of energy transmission paths of the source nodes and introduces the quantitative representation of topology change based on the betweenness centrality. A multi-level deep neural network, which integrates the sub-region state information, topology change information and fault description information, realizes the multi-classification of voltage stability state based on the measured information. Experimental results show that the proposed approach can improve the accuracy of stability state determination obviously. Facing multiple fault states, this study realizes the static voltage stability situational assessment driven by "model+information" from the perspective of qualitative analysis.
(3)An empirical information driven unsupervised voltage stability state evaluation approach is proposed. Aiming at the problem of low modeling accuracy caused by the increasing complexity and randomness of power grid system and component, an approach for dimensionality reduction-based clustering guided by empirical rules is proposed. The optimal latent space for voltage stability state evaluation is constructed based on Siamese autoencoder, and further combined with clustering analysis to realize the effective division of voltage stability level without mechanism modeling. Experimental results show that the proposed approach can effectively approximate the theoretical true value. Based on unsupervised learning, this study explores an effective mode for information driven qualitative static voltage stability situational assessment.
(4)An approach of critical nodes and regions identification based on spatial-temporal information is proposed. The change degree of node voltage phase trajectory is used as the index for node importance and correlation degree measurement. The kinematic and morphological attribute information of voltage phase trajectory is extracted from the aspect of time and space without the need of grid component model and topological structure parameters. The critical nodes and regions are identified dynamically based on a two-stage information fusion and clustering analysis. Experimental results show that the identification results of the proposed method can accurately reflect the changes of voltage stability situation. From the perspective of traceability analysis, this study realizes the information driven search and positioning of critical elements of static voltage stability situation, forming a closed-loop of the overall research.
(5)Integrating the above key technologies, an on-line voltage stability situational assessment prototype verification system for smart grid is designed and implemented. Its overall architecture and core functions are described and the effectiveness of the proposed approaches are verified.

Pages143
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/39253
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
白熹微. 信息驱动的电网静态电压稳定态势评估方法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2020.
Files in This Item:
File Name/Size DocType Version Access License
学位论文.pdf(7000KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[白熹微]'s Articles
Baidu academic
Similar articles in Baidu academic
[白熹微]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[白熹微]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.