Framework for Upscaling Missing Data in Electricity Consumption Datasets Using Generative Adversarial Networks

One of the leading issues in adopting electricity load prediction today is the lack of high-quality and high-resolution real-world datasets. This poses a major problem especially in the context of electricity load prediction where high quality data are essential. To address this issue, this paper presents a framework that transforms datasets with missing values into high quality and high-resolution datasets using Generative Adversarial Networks (GANs). The capability of this framework was exhibited through a case study, the CIC-IPN electricity consumption dataset. Results show that the framework was able to successfully impute the missing values in the dataset while capturing the general patterns in the data. This framework can then be used to upscale other electricity datasets that contain missing values which can then be further used for electricity load prediction for smart cities and smart buildings.