Locational Detection of False Data Injection Attacks in Smart Grids: A Cloud–Edge Framework Based on Split Learning

False data injection attacks (FDIA) pose significant threats to the security of distribution networks, jeopardizing the integrity of measurements and the accuracy of decisionmaking processes. Furthermore, data-sharing concerns present a substantial obstacle to the development of a comprehensive FDIA localization framework. To address these challenges, this paper proposes a cloud-edge framework based on split learning for the localization of FDIA in distribution networks. The proposed framework consists of two primary components: the grid edge computing units at the edge, which includes the local feature extractor and the feature classifier, and the cloud where the global feature extractor is located. The local feature extractors in the grid edge computing units at the nodes are tasked with extracting local features from node measurements. The cloud is responsible for deriving global/spatial features from the local features extracted by each grid edge computing unit distributed throughout the network. Lastly, each feature classifier orchestrates the conclusive processing stage to make decisions at the grid edge based on the spatial features extracted by the cloud. Attention layers namely the self-and the spatial attentions, are utilized to focus on the most important features for the local feature extractors and the global feature extractor, respectively. Data-sharing concerns are mitigated as both the raw measurements and the labels are kept at the grid edge. Comparative studies indicate that the proposed framework maintains strong resilience to varying attack strengths and effectively handles noisy measurements.