CSI-Based temporal segmentation of standardized operations in distribution substations

To address occlusion and lighting constraints in distribution substation monitoring, we leverage Channel State Information (CSI)-based WiFi sensing and propose a continuous action segmentation system for power grid contexts. Built around the 1D-PowerSeqNet temporal network with integrated temporal feature enhancement and adaptive granularity optimization, the system enables direct mapping from raw CSI streams to action boundaries and end-to-end segmentation of variable-length sequences. Validation in a real-world 'hot standby to maintenance' scenario on seven sequential operations, including shaking the handcart, opening the cabinet door, disconnecting the secondary switch, even under strong electromagnetic interference, overcoming the bottleneck of conventional methods that rely on action pauses. This study establishes a technical foundation for non-visual safety monitoring within the Power Internet of Things framework. © Published under licence by IOP Publishing Ltd.