Federated-WDCGAN: A federated smart meter data sharing framework for privacy preservation
Zhiqiang Chen, Jianbin Li, Long Cheng, Xiufeng Liu
Abstract
Energy consumption data are crucial for various smart energy management applications, such as demand forecasting, customer segmentation, and energy efficiency analysis. However, collecting sufficient training data can be a challenge due to data privacy concerns, technical barriers, and cost issues. This paper presents a novel data generation model called Federated-WDCGAN, which combines federated learning with an enhanced generative adversarial network to generate scalable and realistic energy consumption data while preserving privacy. The proposed model is trained in a federated manner, where the generator is trained to generate realistic energy consumption data, and the discriminators are trained to distinguish the generated data from real data. The proposed model is evaluated through analysis of the generated data and its use in a machine learning task for the classification of household characteristics, demonstrating its ability to produce high-quality data comparable to real data in terms of statistics, patterns, and classification performance. The privacy-preserving capabilities of the model are also assessed using non-independently identically distributed and independently identically distributed data, achieving satisfactory results.