MPC-friendly Vertical Federated Learning for Electricity Consumption Prediction in Smart Grids

The increasing complexity of electricity consumption patterns in smart grids requires accurate prediction models to ensure grid stability and efficient resource management. However, the decentralized nature of smart grids, coupled with strict privacy regulations, poses significant challenges to data sharing and collaboration between utility companies. Vertical Federated Learning (VFL) provides a privacy-preserving framework for collaborative model training, where multiple parties can contribute to a shared model without revealing sensitive data. Despite its promise, integrating VFL with secure Multi-Party Computation (MPC) techniques to protect data privacy introduces substantial computational and communication overhead, limiting scalability and efficiency in real-time applications. This paper proposes an MPC-friendly vertical federated learning framework tailored for electricity consumption prediction in smart grids. The framework optimizes MPC protocols to reduce computation and communication costs, making it practical for large-scale, distributed energy systems. By leveraging secure aggregation techniques and efficient model updates, the proposed system preserves privacy while maintaining high prediction accuracy. Experimental results demonstrate that our approach outperforms traditional VFL methods in terms of efficiency and scalability, offering a viable solution for privacy-conscious, real-time electricity consumption forecasting in smart grids.