A topology analysis and genetic algorithm combined approach for power network intentional islanding

Intentional islanding is to determine proper network splitting strategy while ensuring local power balance and transmission capacity constraints when islanding operation is unavoidable. The intentional islanding problem is very complicated in general because a combinatorial exploitation of strategy space is required. This paper apply a topology analysis and genetic algorithm combined approach for determining proper splitting strategies of large-scale power networks. Topology analysis is used to simplify the original power network into a simple equivalent network so that the splitting strategy space world be dramatically reduced; while the genetic algorithm incorporated with the breadth-first search (BFS) is employed to determine the final proper splitting strategy in the simplified power network. Two additional applications, mimicking weak connections between islands and obtaining specific pre-defined islands, of the proposed method are introduced. Simulation results on several test system show that the proposed approach can quickly provide proper splitting strategies and is effective for larger-scale power systems.     

基于Tri-training-SSAE半监督学习算法的电力系统暂态稳定评估

基于机器学习的暂态稳定评估方法主要采用监督学习方法,为了解决监督学习方法所需的有标签样本难以获取的问题,提出基于三体训练-稀疏堆叠自动编码器(Tri-training-SSAE)半监督学习算法的电力系统暂态稳定评估方法。构建基于堆叠稀疏自动编码器的暂态稳定评估模型;在传统的三体训练过程中加入伪标签样本置信度判断,以减小噪声数据对模型训练的影响;以堆叠稀疏自动编码器为基分类器构建三体训练-稀疏堆叠自动编码器模型,利用大量的无标签样本提高模型的泛化能力。通过IEEE 39节点系统与华东某省级电网进行分析验证,结果表明,所提方法在有标签样本数较少时具有更高的评估准确度。     

考虑同调约束的电力系统主动解列断面的搜索方法

电力系统主动解列是避免发生大停电事故和系统崩溃的有效保护措施,需要及时、准确地确定解列方案。文中提出一种考虑发电机同调约束的最佳解列断面的搜索方法。首先基于故障后广域测量系统(WAMS)提供的实时动态响应信息,利用改进拉普拉斯特征映射算法在线识别同调机群,利用发电机同调虚拟节点聚合同调机群,减小解列断面的搜索空间,然后利用加权拉普拉斯分群策略得到初始的解列断面。最后考虑断面有功功率的方向,利用启发式变邻域搜索算法确定最佳的解列断面,使得解列后各孤岛的净不平衡功率最小。文中方法能充分考虑不同故障模式所带来的同调行为的差异性,使得主动解列控制具有选择性和鲁棒性。通过新英格兰39节点系统验证了所述方法的有效性和适应性。     

Performance and suitability assessment of controlled islanding methods for online WAMPAC application

Controlled islanding can be used as an efficient action to prevent catastrophic blackouts. It splits a power system into a group of stable islands, according to an islanding solution that has been generated using an islanding method. A detailed understanding of these methods is a sound basis for the development of feasible controlled islanding methods for future networks. In this paper three methods (a) Ordered Binary Decision Diagram, (b) Weak Connection and (c) Spectral Clustering, are selected as typical islanding methods. The paper presents an analysis of their essential differences and their potential suitability for use as part of a practical implementation of controlled islanding. They are critically assessed and contrasted in terms of the objective function, graph model, solving algorithm and time complexity. The key difference between these methods is the nature of the objective function used: minimal power imbalance, minimal dynamic coupling or minimal power-flow disruption. A different objective function may require a different graph model and a different solution algorithm, which may result in the islanding methods having different time complexities and produce different islanding boundaries for the same network. Each method is applied to the IEEE 118-bus system to explore their strengths and weaknesses. Criteria for the selection of a suitable, practical method are discussed, as is the feasibility of Controlled Islanding as a WAMPAC application.     

考虑柔性直流落点约束的最优主动解列断面搜索算法

当电力系统遭受严重扰动后,主动解列策略是防止灾害性大停电事故积极而有效的手段,其核心内容是搜索最优解列断面,通常理解为是一种含约束的组合优化问题。提出了一种基于谱聚类的考虑柔性直流落点约束的最优主动解列断面搜索算法,该方法通过谱聚类算法将电力系统的最优解列断面搜索问题映射为图的分割问题进行求解,直接求取最优解且计算速度较快;以最小有功潮流冲击为目标函数,最小化解列后系统潮流的改变提高了孤岛系统的暂态稳定性;考虑了基于电压源换流器的高压直流输电(VSC-HVDC)的终端落点作为约束条件,将VSC-HVDC的终端置于不同孤岛内可实现孤岛间功率的交换,降低了系统整体恢复稳定的成本。最后通过IEEE 10机39节点系统以及厦门系统实际算例的仿真验证,证明了该方法的有效性和快速性。     

电力系统主动解列断面搜索方法与孤岛调整策略

主动解列是一种基于电力系统实时动态信息的紧急控制手段。及时而恰当的主动解列可以阻隔故障传播,避免保护连锁动作可能导致的系统崩溃,有利于大扰动后电力系统的快速恢复。在此背景下,提出一种包括快速搜索解列断面和优化调整孤岛的系统主动解列方案。首先,对在机器学习领域发展起来的谱聚类算法进行改进,提出含约束谱聚类算法,以计及发电机组的同调约束,从而将解列断面搜索问题转化为广义特征值求解问题。为克服在含约束谱聚类算法中采用传统k-medoids算法存在的对初始中心点敏感、搜索效率低的缺点,提出改进k-medoids算法并将其与约束谱聚类算法相结合,以求取最优解列断面。然后,对于解列后每个不满足安全约束的孤岛,优化调整其发电机组出力,必要时也可削减一些负荷,以维持每个孤岛的安全运行。最后,以IEEE 118节点系统和实际电网为例,说明所提方法的可行性和有效性。     

考虑同调机组的电力系统主动解列断面搜索方法*

将电力系统解列断面搜索问题转化为单目标多约束优化问题,提出了一种考虑同调机组的解列断面搜索新方法。节点间电抗值的大小反映了其电气连接强弱程度。首先以节点间线路电抗值为线路权值,利用Dijkstra算法计算同调机群中发电机之间最小电抗连接线路,得到同调机组连通图,之后用prim算法得到同调机组连通图的最小生成树,并将其等值为一个节点,从而保证电力孤岛内机组连通性。最后利用Dinic算法求得非同调机群间最小潮流割集,得到最优解列断面。对经典IEEE-39节点系统进行算例仿真,验证了文中所提出方法能够快速准确定位解列断面。     

基于贪心算法的自适应解列断面自动搜索方法：(一)解列指标体系与算法设计

现有的解列断面搜索方法仅能考虑发电机同调、孤岛功率平衡等少数约束条件,在大系统的解列断面搜索过程中需要事先人工化简系统,无法实现从在线调度数据出发自动获得解列断面。文中以交直流系统的自适应解列控制为研究目标,提出了交直流电网解列指标体系,并设计了基于贪心算法的解列断面自动搜索方法。该方法根据系统发电机慢同调分群结果,确定解列孤岛数量及孤岛初始搜索节点,然后,对各孤岛一级相连节点进行评分,确定应该并入各孤岛的节点;最后,设计了解列策略校验环节。     

基于拓扑连通性约束遗传算法的主动解列断面搜索

随着大规模新能源互联电网的发展,电力系统发生连锁故障的风险越来越高,适当的主动解列可以阻止故障的传播。为解决主动解列断面求解过程中的系统拓扑连通性和计算复杂度高的问题,提出了一种含拓扑连通性约束遗传算法的主动解列断面搜索方法。该方法首先构建了解列断面搜索的数学模型,然后基于图论知识提出了一种系统拓扑连通性约束,并加入到遗传算法中。同时提出了一种系统拓扑简化和预处理方法。最后将简化后的系统拓扑和数据输入到含拓扑连通性约束的遗传算法中进行主动解列断面的求解。IEEE 118节点系统算例表明,所提算法能够保证求解出的主动解列断面具有拓扑连通性且有功功率基本平衡,系统拓扑简化和预处理方法能够有效提高算法的运行效率。     

Islanding detection using wavelet transform and rate of change of frequency relay method in presence of different distributed generation technologies

In this paper, islanding detection in the presence of different technologies of distributed generations (DGs) is discussed. Three types of DGs, namely a synchronous generator (SG), a wind energy conversion system (WECS), and a photovoltaic (PV) system, with distribution network parameters as well as linear and nonlinear loads are simulated using the Power System Computer Aided Design/Electromagnetic Transients Including DC (PSCAD/EMTDC) software. To detect the island state, a comprehensive islanding detection algorithm based on wavelet transform (WT) and rate of change of frequency (ROCOF) relay method is considered. To determine the islanding threshold value, the voltage and frequency values of sensitive buses are sampled at regular intervals, and the signals are analyzed with WT using MATLAB. The proposed algorithm could detect the exact island state definitively with the consideration of various disturbance signals such as capacitor bank switching, induction motor switching, and the sudden disconnection of power sources and loads from different buses of the distribution network. The results of the simulation are validated with the logic of ROCOF relay islanding detection method, which determines the accuracy threshold value when the power generation and load consumption difference is negligible. In addition, the proposed algorithm based on the rested DG's output power in the islanding state can supply prioritized loads based on thier importance. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

Power system with multi-type FACTS devices states estimation based on predictor–corrector interior point algorithm

This paper addresses a state estimation problem of power systems incorporating various flexible alternating current transmission system (FACTS) devices. Static var compensators, thyristor controlled series compensators, and unified power flow controllers are considered since they represent various kinds of FACTS devices. By considering constraints of system buses and devices, the estimation problem can be formulated as a nonlinear optimization with constraints. An algorithm based on predictor–corrector interior point is applied to solve the problem. The IEEE 14- and 118-bus systems modified by incorporating FACTS devices are used as test systems to verify and demonstrate the effectiveness of the proposed algorithm. Numerical results indicate that the proposed algorithm performs better than the primal dual interior point based method even the presence of bad measurement data.     

一种电力系统失步解列面的实时搜索方法

为故障后失去稳定的电力系统确定解列断面以平息系统振荡可以防止系统完全崩溃,如何快速寻找不同失稳模式下的解列面成为解列控制的关键问题之一。提出电力系统失步解列面实时搜索方法,该方法的优化目标为解列后各孤岛内发电机和负荷不平衡尽可能小。方法包含3个阶段:基于潮流追踪的系统分区、基于发电机分群信息的解列面初步确定和基于初步解列断面的解列断面改善。算法具有时间复杂度小、计算速度快的特点。基于C++语言编制实时搜索程序,对IEEE118节点系统进行解列面搜索仿真计算,仿真算例验证了方法的有效性和快速性。     

多层次结构与半监督学习的谣言检测研究

当前谣言检测工作主要基于监督学习,需要人为标记数据而导致检测具有滞后性。为了充分利用大量的未标记数据,及时检测社交网络中的虚假谣言。提出了一种基于多层次结构与半监督学习谣言检测模型(multi-level semi spuervised graph convolutional neural network, MSGCN)。该模型构建了一种多层次检测模块,基于图卷积网络对有限的标记样本进行训练以提取多层次传播结构特征、扩散结构特征和全局结构特征。其次,引入随机模型扰动集成无标签数据的动态输出进行一致性预测,提出互补伪标签法来获取高质量伪标签数据,并将其加入标记数据扩充样本。最后在有监督交叉熵损失和无监督一致性损失约束下提高模型质量。在公开的Twitter15、Twitter16和Weibo数据集上的实验结果表明,所提出模型在30%标记样本下准确率达到88.3%、90.1%和95.5%,在少量的标记样本下便可达到优异的成绩。     

Improved multilevel bipartitioning for controlled power system islanding

Nowadays interconnected power systems work close to their limits because of increased economic benefits. If a severe disturbance occurs in any part of the interconnected power system, it can be propagated through it. If no intelligent supervisory action is taken, a power system may be driven into an emergency state which could either cause system collapse by natural islanding or total system blackout. One possible measure to be put in effect in an emergency state is controlled power system islanding. In this paper, our previously proposed multilevel approach for controlled islanding is improved by using slow coherency grouping of generators in the aggregation phase. The approach has been applied to the IEEE 118 bus power system. The results indicate that the best partitioning of the power system is obtained with minimum generation/load imbalance in the islands, and with the coherent groups of generators maintained in each island making the restoration easy. A comparison of the computational complexities involved in the spectral, multilevel kernel k‐means and the improved multilevel approaches is made by applying these methods to the IEEE 9 bus, the IEEE 30 bus, the IEEE 39 bus, the IEEE 57 bus, the IEEE 118 bus, the IEEE 300 bus, the 2383 bus Polish and the 2746 bus Polish power systems to show the suitability of the proposed approach for controlled islanding of large‐scale power systems. © 2011 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

数据驱动的同调机组分群及主动解列技术

新型电力系统发生区域性故障时系统的安全稳定性问题日益突出,主动解列作为防止故障扩散的有效手段在系统稳定研究中逐渐获得关注。从同调机组分群技术和解列断面搜索技术两个方面对主动解列开展研究,结合时频分析、特征提取和图论等方法,提出基于改进聚类算法的同调机组分群技术和基于多层图分割的解列断面搜索技术,并在IEEE-39节点标准测试系统中验证了所提方法的适用性。     

基于有向图的含分布式电源配电网孤岛划分模型

含分布式电源配电网的孤岛运行模式可以充分提高分布式电源利用率和配电系统供电可靠性。文中提出一种基于有向图模型的含分布式电源配电网的孤岛划分新模型。首先,提出一种基于有向图的配电网结构模型,该模型通过引入"虚拟节点"和"虚拟需求"建立了支路方向可变的配电网模型,从网络结构模型上保证了配电网孤岛运行区域的连通性和辐射状结构。然后,基于有向图模型,建立了考虑联络开关可操作性、负荷重要等级、负荷可控性以及分布式电源运行特性的孤岛划分模型,并将其转化成一个混合整数线性规划问题以便于求解。最后,采用PGE 69节点配电系统进行测试,验证了所提出的孤岛划分策略的有效性。     

基于对角加边模型的多区域无功优化分解算法

基于对角加边矩阵结构,提出了一种新的多区域电力系统离散无功优化分解算法.该方法先将电力系统按照一定的规则进行分区,通过引入虚拟节点构成各区域之间的边界网络,并采用内嵌离散惩罚的非线性原对偶内点法求解.最终所形成的线性修正方程组的系数矩阵具有对角加边结构.由此提出2种分解方法实现各区域修正方程的独立求解,寻找全系统及其各区域的近最优离散解.以IEEE 118节点试验系统和2个实际系统(538节点和1133节点系统)作为算例,通过对集中优化方法和2种分解方法进行比较分析验证了所提出方法的有效性.     

基于CSP的复杂配电网孤岛划分方案研究

在含分布式电源的复杂配电网系统发生故障时,制定合理的孤岛划分方案可提高系统的供电可靠性。利用复杂配电网简化模型,采用两阶段策略解决孤岛划分问题。首先提出了基于约束满足问题的孤岛划分方法,离线生成满足约束问题的单元集合,将孤岛划分问题转化为寻求孤岛最优解的问题。然后在配电系统发生故障时,采用启发式的回溯搜索策略,得出最终的孤岛划分方案。通过对IEEE69节点配电系统进行仿真．验证了该方法有效性和实用性。     

基于起振特性的强迫振荡扰动源定位及解列方案

强迫振荡已成为目前低频振荡事故的主要形式,快速定位并解列扰动源是抑制强迫振荡的有效手段。在起振阶段,由于惯性环节的存在,离扰动源较远的节点测量量振荡相位较之离扰动点较近的节点滞后,此特性可作为扰动源定位能量函数法的辅助判据,用于从多个疑似扰动源所在区域中正确定位扰动源。通过对南方电网某方式的仿真验证了该方法的有效性。结合强迫振荡扰动源解列的特点,提出两种解列方案,分别为:基于电网实际分区,解列扰动源所在的区域;基于图论提出适用于扰动源解列的断面调整规则,确定解列断面。     

分布式发电条件下的配电网孤岛划分算法

在分布式发电条件下,孤岛可作为提高配电网供电可靠性的一种重要运行方式。为在配电网发生故障时及时制定出优化的孤岛划分方案,提出了孤岛划分算法。该算法利用配电网简化模型,根据孤岛运行时的功率平衡要求以及各种负荷对供电可靠性的不同要求,采取了启发式的搜索策略, 能够在短时间内得到可行的孤岛划分方案。文中还给出了2 个典型的孤岛划分实例。理论和实例分析表明,该算法可在配电网发生故障以后,根据故障前的实际运行情况动态地生成优化的孤岛划分方案。