基于粗糙集和Bi-LSTM-Attention网络的电力系统暂态评估

为了降低不稳定因素对现代复杂的电力系统暂态评估的影响,在基于模糊领域单参数粗糙集和注意力机制的双向长短期记忆(Rs-Bi-LSTM-Attention)网络下,提出了一种电力系统暂态稳定评估模型。该模型首先用模糊领域单参数粗糙集对电力系统潮流数据进行属性约简,该方法既保证了准确率又提高了核心因素的权重;然后,通过Bi-LSTM-Attention网络对约简后的数据与电力系统暂态稳定状态之间建立映射关系,其中引入了LayerNormalization对高层神经网络的输入数据进行处理,使得高层神经网络不仅能适应低层参数更新,而且可以加快网络的收敛速率。最后,用评估准确率和F_(1-measure)两种评价指标对该模型的性能进行评估。IEEE39算例分析表明,Rs-Bi-LSTM-Attention模型比机器学习模型和部分深度学习模型具有更高的优越性。

基于改进胶囊网络的电力系统暂态稳定评估

针对在电力系统发生故障后,快速、准确评估电网暂态稳定性的需求,提出了一种基于改进胶囊网络和短时受扰轨迹的电力系统暂态稳定评估方法。该方法通过仿真得到量测数据,利用胶囊网络的特征提取能力挖掘隐含在输入数据中的关键特征,建立输入轨迹与电网稳定与否的映射关系,评估系统的暂态稳定性,降低了网络的错判样本,有效提升了模型评估的准确率,增强了模型的鲁棒性以及泛化能力。仿真验证了该方法的可行性,并且所提方法相比于其他机器学习方法的评估模型更加优越。

基于残差时空图神经网络的电力系统暂态稳定评估

针对传统数据驱动方法在用于电力系统暂态稳定评估时,由于未考虑到系统的时空变化特性而导致泛化能力不强的问题,提出了一种引入注意力机制的残差时空图神经网络方法。将电力系统的网络拓扑结构与电力系统量测数据相结合,引入时间注意力和空间注意力机制,从而提高了模型的泛化能力。为了解决训练过程中存在的梯度消失问题,引入残差结构,同时加快了训练速度。以10机39节点系统为算例进行仿真验证,结果表明,该模型能在部分节点未配置同步相量测量单元的情况下获得较好的评估结果;通过适当选取故障时刻采样数据,可以进一步提高模型的评估性能。

基于耦合模拟退火和最小二乘支持向量机的电力系统暂态稳定评估

针对现有支持向量机(SVM)在评估精度上的不足,提出一种基于耦合模拟退火和最小二乘支持向量机(LSSVM)的电力系统暂态稳定评估方法,将SVM中的约束转化为等式约束,经验损失用损失函数表征,实现二次规划与线性方程问题的转化。采用系统指标造原始特征集,通过最大相关最小冗余特征选择方法进行特征压缩,确定强相关特征集,将特征子集映射到高维空间,使非线性与线性分类问题的转化;然后,用耦合模拟退火法确定LSSVM的最优参数提高计算精度,并进行暂态稳定判别。最后,采用新英格兰10机39节点系统和某实际电网系统验证所提方法的有效性和准确性。

基于核主成分分析和深度置信网络的暂态稳定评估

针对电力系统暂态稳定评估实时性较差以及错误率较高的问题,提出了一种核主成分分析结合深度置信网络的暂态稳定评估方法。首先,构造了一组反映电力系统暂态稳定的特征向量;然后,基于核主成分分析法对特征向量集进行特征提取,降维特征向量维数以及过滤冗余特征,将降维后的特征向量传输至深度置信网络;最后,进行训练分析,训练过程包括预训练和微调,优化网络参数,提升深度置信网络评估精度。新英格兰10机39节点系统仿真结果表明,该方法可以有效降低输入数据的维数,去除冗余特征,降低暂态稳定性评估的错误率和测试时间,能准确、快速地判断电力系统的稳态状态。

基于特征量和卷积神经网络的暂态电压稳定评估

针对目前传统方法无法准确快速判断系统暂态电压稳定性的问题,提出一种基于特征量和卷积神经网络(convolutional neural networks,CNN)的暂态电压稳定评估方法.首先基于传统电压稳定指标分解及响应数据,从原始电气量中进行36维电压稳定构建特征量,将特征量输入卷积神经网络中进行有监督学习,最后将训练所得到的模型应用于电力系统暂态稳定评估中.利用新英格兰10机39节点标准算例对电网进行了仿真,结果表明该方法具有错误率低、精确率高、测试时间短的特点,能够准确快速判断出系统的暂态电压失稳,提高了电力系统的稳定性.

Transient Stability Assessment of Synchronous Generator in Power System with High-Penetration Photovoltaics （Part 2）

In the previous paper [1], the transient stability of synchronous generator in power system with high-penetration PV （photovoltaic） was assessed by simulation analysis of a single-machine infinite-bus system model. Through the simulation analysis, we have obtained some conclusions in terms of the impact of high-penetration PV on the stability. However, for more accurate assessment of the transient stability, it is necessary to analyze various simulation models considering many other power system conditions. This paper presents the results of the analysis for the transient stability simulation performed for IEEE 9-bus system model, in which the effects of various conditions, such as variety of power sources （inverter or rotational machine）, load characteristics, existence of LVRT （low-voltage ride-through） capability and fault locations, on the transient stability are investigated.

Transient stability probability of a power system incorporating a wind farm

An analytical approach for probabilistic evaluation of transient stability of a power system incorporating a wind farm is presented in this study. Based on the fact that the boundary of practical dynamic security region(PDSR) of a power system with double fed induction generators(DFIG) can be approximated by one or few hyper-planes in nodal power injection space, transient stability criterion for given configurations of pre-fault, fault-on and post-fault of a power system is to be expressed by certain expressions of linear combination of nodal injection vector and the transient stability probability(TSP) is further obtained with a much more simplified expression than the complex integral. Furthermore, considering uncertainties of nodal injection power including wind power and load, TSP is calculated analytically by Cornish-Fisher expansion, which can provide reliable evaluation results with high accuracy and much less computing time compared with Monte Carlo simulation. TSP and its visualization can further help operators and planners be aware of the degree of stability or instability and find critical components to monitor and reinforce. Test results on the New England 10-generators and 39-buses power system show the method's effectiveness and significance for probabilistic security assessment.