Data-driven and privacy-preserving risk assessment method based on federated learning for smart grids

Abstract

Timely and precise security risk evaluation is essential for optimal operational planning, threat detection, and the reliable operation of smart grid. The smart grid can integrate extensive high-dimensional operational data. However, conventional risk assessment techniques often struggle with managing such data volumes. Moreover, many methods use centralized evaluation, potentially neglecting privacy issues. Additionally, Power grid operators are often reluctant to share sensitive risk-related data due to privacy concerns. Here we introduce a data-driven and privacy-preserving risk assessment method that safeguards Power grid operators’ data privacy by integrating deep learning and secure encryption in a federated learning framework. The method involves: (1) developing a two-tier risk indicator system and an expanded dataset; (2) using a deep convolutional neural network -based model to analyze the relationship between system variables and risk levels; and (3) creating a secure, federated risk assessment protocol with homomorphic encryption to protect model parameters during training. Experiments on IEEE 14-bus and IEEE 118-bus systems show that our approach ensures high assessment accuracy and data privacy.