基于时空图卷积网络的谐波状态估计方法研究

Harmonic State Estimation Based on Spatiotemporal Graph Convolutional Network

随着现代电力系统中分布式新能源的广泛接入,谐波问题愈加复杂。现有谐波状态估计方法存在估计精度低、电网参数难以获取、缺乏系统性等弊端。首先,该文对谐波数据时空特性进行分析,基于数据驱动融合图卷积神经网络和门控循环单元对未知节点谐波状态进行估算;其次,提出子图分割方法,将整个系统划分为若干子图独立进行数据采集和状态估计,合并后实现了谐波状态全网可观性,解决了监测装置数量不足的问题;最后,仿真算例数据和实际量测数据均验证了方法的有效性和适用性,为谐波状态估计问题提供了新的解决方案。

With the widespread access of the distributed new energy sources in the modern power systems,the problem of harmonics has become more complex.The existing harmonic state estimations have the disadvantages of low estimation accuracy,difficult perception of the power grid parameters,and lack of systematicity.Firstly,this paper analyzes the spatiotemporal characteristics of the harmonic data,and estimates the harmonic state of unknown nodes based on the data-driven fusion graph convolutional neural network and the gated recurrent unit;secondly,a subgraph segmentation is proposed to divide the whole system into several independent subgraphs to indipendently carry out the data acquisition and state estimation.The whole network observability of the harmonic state is realized after the merger of those subgraphs,which solves the problem of having insufficient monitoring devices.Finally,the effectiveness and applicability of the method were verified based on simulation case data and actual measurement data,providing a new solution for the problem of harmonic state estimation.