Voltage Sag State Estimation Based On Improved Generative Adversarial Network

Hongze Qu, Linhai Qi, Hong Wang

Abstract: State estimation is an effective means to solve the observability of the whole network. Traditional voltage sag state estimation usually uses the combination of electrical environmental parameters and mechanism analysis to construct mathematical models. Due to the high permeability, dispersion, and dynamic time-varying nature of distributed power sources and nonlinear loads, power quality operation scenarios are more complicated and model accuracy is reduced. Data-driven state estimation can avoid the limitations of physical models based on mechanism analysis. In this paper, a deep learning method using unsupervised loop generation of confrontation is proposed to realize voltage sag state estimation. This method does not require prior knowledge and calculation of electrical topology relationships, and uses the coupling relationship between multiple nodes for state estimation, effectively solving the influence of the dynamic characteristics of the grid environment on the coupling relationship caused by factors such as line adjustment, grid parameter changes, and new energy uncertainties., Improve the generalization ability of the model. The paper tests the model under the environment with noise, and verifies the validity of the model.

Keywords: cycle generative countermeasure network, multi-node state estimation, voltage sag, network-wide visibility