Intelligent Interactive Systems for Power Grid Monitoring Based on Natural Language Processing
Abstract
High-voltage transmission grids are experiencing increasing operational complexity due to alarm floods, renewable energy variability, and reduced workforce. Manual processing of massive supervisory control and data acquisition (SCADA) messages is becoming impractical. This paper systematically reviews how Natural Language Processing (NLP), including intent recognition, knowledge graphs, and dialogue management, can develop Intelligent Interactive Systems (IIS) to efficiently filter alarms, respond to operator queries, and propose corrective measures in real time. We examine current monitoring workflows for 220 kV and above grids, analyze critical NLP methodologies tailored for the power systems domain, propose an IIS architecture with key functional modules such as automated fault information extraction, alarm prioritization, and decision support, and present relevant performance metrics. Four representative field prototypes demonstrate a significant enhancement in alarm event recognition (up to $96 \%$ accuracy) and a substantial reduction in operator workload (up to $50 \%$). Finally, open research challenges, including few-shot learning, edge computing security, and system explainability, are identified, indicating paths for future development.