Incentive edge-based federated learning for false data injection attack detection on power grid state estimation: A novel mechanism design approach

With the growing concern in security and privacy of smart grid, false data injection attack detection on power grid state estimation now faces new challenges including unknown system parameters and small decentralized data sets with strategic data owners. To deal with these technical bottlenecks, this paper proposes a novel edge-based federated learning framework for false data injection attack detection on power grid state estimation, which has great potential in real-world applications with unknown system parameters. Furthermore, to seek a high detection accuracy with small decentralized data set and strategic data owners, an incentive mechanism is designed to encourage the desired data owners contributing to false data injection attack detection. To explore the impact of the incentive mechanism on the detection accuracy, a bi-level model depicting the data owners’ participation in false data injection attack detection is formulated, based on which the impact is quantified. Moreover, a novel preference criterion is proposed for optimal mechanism design. It can achieve the optimal detection accuracy under a certain incentive budget. The incentive mechanism is designed and tested for 100 Monte Carlo scenarios. Simulations of false data injection attack detection on power grid state estimation show that the proposed framework outperforms the existing works without mechanism design.