基于消息传递图神经网络的电力系统状态估计

State Estimation of Power System Based on a Message Passing Neural Network

近年来,随着电网规模日益增大,系统拓扑变化愈加频繁,拓扑变化组合向多样化趋势发展。然而,现有数据驱动状态估计模型仅能处理欧式数据,不能有效挖掘像拓扑信息这样的非欧式数据,因此现有数据驱动模型在拓扑频繁变化时的适应性较差。该文提出一种基于消息传递图神经网络(messagepassingneural network, MPNN)的电力系统状态估计模型。首先,利用拓扑参数和量测信息构建图数据集;其次,基于不同拓扑下的图数据训练消息传递图神经网络,得到状态估计模型;最后,在线应用时将该拓扑下的图数据输入已训练好的网络模型即可得到当前断面的状态量。通过对IEEE标准系统和中国某实际省网的算例测试,并将估计结果与加权最小二乘法、加权最小绝对值法以及深度神经网络算法和卷积神经网络算法进行比较。结果表明,该算法更能适应大规模电网中实时拓扑变化的特性。

In recent years,data-driven methods have been widely used in power system state estimation.However,the existing data-driven state estimation models can only deal with Euclidian data and cannot effectively mine non-Euclidian data such as topology information.Therefore,the existing data-driven state estimation models have poor adaptability when the topology changes frequently.This paper proposes a power system state estimation model based on Message Passing Neural Network(MPNN).Firstly,topology information and measurement information are used to construct graph data sets.Secondly,the state estimation model is obtained by training the message passing graph neural network based on graph data in different topologies.Finally,the state quantity of the current section can be obtained by inputting the graph data under the topology into the trained network model during online application.The results are compared with the weighted least square method,the weighted least absolute value method,the deep neural network algorithm and the convolutional neural network algorithm.The results show that the algorithm can better adapt to the time-varying characteristics of power grid topology.