Augmented Generative Adversarial Network-Based Method for Harmonic State Estimation

Traditional harmonic state estimation methods are limited by few measurement devices, difficulty in obtaining accurate harmonic impedance, complex network topology, and changes in grid operation. These factors cause problems such as underdetermined measurement equations, non-global observable systems, and difficulty in extracting accurate coupling relationships between buses. This paper proposes a novel method based on an augmented generative adversarial network for harmonic state estimation. The efficient extraction of electrical image features by the neural network method is achieved by transforming the temporal sequence data into electrical images. The generator based on deep residual networks and the improved residual block structure are used to improve the feature learning capability of the generator. In addition, the loss function of the generator takes into account the difference between the real samples and the generated samples in the high-frequency component. The validity and accuracy of the proposed method have been verified by simulation analysis.