Abstract: To address the long control path, backbone-network congestion caused by centralized data transmission, and center-dependent fault diagnosis in conventional PLC-based railway traction substations, a distributed collaborative scheme using edge-computing PLCs is proposed. A three-layer architecture consisting of field edge nodes, a station-level edge gateway, and the dispatching center is established. Deterministic protection and interlocking, data preprocessing, and preliminary fault identification are deployed close to the equipment. Data classification, event-triggered transmission, cross-bay coordinated control, model version management, and security isolation are introduced to decouple real-time control from intelligent analytics. A retrofit project in a high-speed railway traction substation was evaluated through continuous field tests. The average control latency reached 18 ms, the average backbone bandwidth utilization decreased from 72% to 12%, and fault-location time was reduced from approximately 6 min to 8 s. A single edge-node outage did not affect the normal operation of other bays. The results indicate that integrating edge computing with PLC technology can improve real-time performance, communication efficiency, and fault-response capability without changing the existing protection safety boundary, providing an engineering reference for digital upgrading of traction substations.