Edge-cloud collaborative large model for autonomous operation and maintenance in distribution substations

With the increasing complexity of distribution substations driven by distributed generation, environmental variability, and aging infrastructure, traditional centralized monitoring systems struggle to meet requirements for real-time decision-making, risk prediction, and autonomous operation. This paper proposes Edge-Cloud Collaborative Large Language Model (EC-LLM), an Edge-Cloud Collaborative Large Model architecture specifically designed to bridge the gap between low-latency edge perception and high-level cloud reasoning. Leveraging a fictional but industry-realistic testbed comprising 12 substations, 850 multimodal sensors, 6,200 hours of image and Supervisory Control and Data Acquisition (SCADA) data, and 21,000 annotated abnormal events, we present a rigorous validation of the framework. Experimental results demonstrate that EC-LLM reduces alarm response latency by 42% (to 290 ms), improves fault localization accuracy by 31%, Furthermore, the system lowers on-site human intervention by 55% compared to conventional centralized architectures. These findings validate that collaborative intelligence—combining lightweight edge perception with domain-enhanced cloud reasoning—significantly enhances the reliability, responsiveness, and autonomy of distribution substation Operation and Maintenance (O and M).