基于生成对抗网络与局部电流相量的配电网拓扑鲁棒辨识

Robust Identification for Distribution Network Topology Based on Generative Adversarial Network and Partial Current Phasor

有源配电网运行方式切换频繁且其量测受噪声影响,要求其拓扑辨识具有实时性和鲁棒性,但传统方法驱动的有限观测条件下拓扑辨识是非确定性多项式(NP)难问题。文中提出一种基于生成对抗网络与局部电流相量的配电网拓扑鲁棒辨识方法。为解决传统监督学习在未知拓扑辨识任务下泛化能力差的问题,通过梯度惩罚优化的条件生成对抗网络(CGAN)学习由线路电流幅值、相角和节点负荷伪测量映射的拓扑分布。同时,以一维卷积神经网络构建生成器,有效利用连续观测窗中时序电流数据,增强算法的抗噪和抗数据缺失性能。此外,仅局部的电流信息需求大幅降低了配电网可观性改造投资。最后,通过算例验证了所提方法的有效性。

The operation mode of active distribution network is frequently switched and its measurement is affected by noise, so its topology identification is required to be real-time and robust. However, the topology identification under limited observation driven by traditional methods is a non-deterministic polynomial(NP)-hard problem. This paper proposes a robust identification method for distribution network topology based on generative adversarial network and partial current phasors. In order to solve the problem of poor generalization ability of traditional supervised learning in unknown topology identification tasks, the conditional generative adversarial network with gradient penalty optimization is used to learn the topology distribution mapped by line current amplitude,phase angle and node load pseudo-measurement. Meanwhile, by constructing generator with one-dimensional convolutional neural network, the sequential current data in continuous observation window are effectively used, which enhances the anti-noise and anti-data loss performance of the algorithm. In addition, the need for only partial current information greatly reduces the investment in observability retrofit of distribution network. Finally, the effectiveness of the proposed method is verified by arithmetic examples.