考虑暂态稳定过程的电力系统运行状态辨识

Power System Operation State Identification Considering the Process of Transient Stability

数据驱动背景下的电力系统状态辨识包含故障前和故障后两个场景,分别具有安全域和稳定域概念下的分析特点。受限于样本特征及应用场景的不一致,现有研究通常将其作为两个独立的问题进行建模。一方面,独立建模忽略了两者间的时序耦合性;另一方面,电力系统运行状态时刻在发生变化,基于单一场景建立的网络模型仅适配当前场景,基于双场景独立建模存在模型切换复杂及参数更新耗时过长等问题。针对以上问题,该文提出一种基于多尺度密集网络的电力系统状态辨识方法,同时适用于故障前和故障后暂态稳定场景。首先,采用自适应池化层结构,建立适用于时间断面特征和时序特征类型输入的电力系统运行状态辨识模型,可同时应用于故障前和故障后的场景;其次,考虑到安全域和稳定域概念下暂态稳定数学模型的时序相关性,设计一种基于知识推理的样本自适应选择机制,通过网络“动态”特性表达故障前和故障后场景间的时序关系,提高计算效率。最后,在新英格兰10机39节点算例系统和实际电网中验证了所提方法的有效性与优越性。

The power system state identification in the data-driven background includes two scenarios,pre-fault and post-fault,which have the analysis characteristics under the concepts of safety domain and stability domain,respectively.Limited by the inconsistency of sample characteristics and application scenarios,existing studies usually model it as two separate problems.On the one hand,independent modeling ignores the coupling between the two.On the other hand,the operating state of the power system is changing all the time;the network model established based on a single scenario is only suitable for the current scenario;independent modeling based on two scenarios has problems such as complex model switching and long time for parameter update.To solve this problem,this paper proposes a power system state identification method based on multi-scale dense network,which is suitable for both pre-fault and post-fault transient stabilization scenarios.First,the adaptive pooling layer structure is used to establish a power system operation status identification model suitable for the input of time section features and time series feature types,which can be applied to both pre-fault and post-fault scenarios.Then,considering the temporal correlation of transient stable mathematical models under the concepts of safe domain and stable domain,an adaptive selection mechanism of samples based on knowledge reasoning is designed to express the temporal relationship between pre-fault and post-fault scenarios through the"dynamic"characteristics of the network,so as to improve the computational efficiency.Finally,the effectiveness and superiority of the proposed method are verified in the New England 10-machine 39-node study system and the actual power grid.