A Power Data Reconstruction Method Based on Super-Resolution Generative Adversarial Network

Abstract

The smart grid is rapidly developing to become highly connected and automated. These advancements have been mainly attributed to the ubiquitous data communication in the grid. However, low sampling frequency will limit the utilization degree of data because low frequency measurement power data contains little information. The existing methods of reconstructing the low-frequency sampling data into the high-frequency sampling data have poor accuracy of data reconstruction since most of them failed to capture the characteristics of power data. This paper proposes a novel method based on super-resolution generative adversarial network (SRGAN) to address this issue. First, we convert power data into data-images. Furthermore, the data-images are used to train the SRGAN model. Finally, the trained generator can be used to reconstruct the low-frequency sampling data into the high-frequency sampling data. Numerical experiments have been carried out based on photovoltaic (PV) power generation time-series data from the State Grid Corporation of China with separately reconstruction of the irradiance and PV power datas. The results demonstrate the superior performance of the proposed method compared with a series of state-of-the-art methods.