Adaptable semantic interoperability in heterogeneous smart grids using Large Language Models

The rapid deployment of Distributed Energy Resources (DERs) has intensified interoperability challenges caused by unregulated technologies, diverse communication standards, and outdated standards that lag behind evolving data architectures. This work proposes a benchmark LLM-based Interoperability Broker (IB) that enables scalable, adaptable, and technology-agnostic communication in smart grids. The IB supports forward and backward compatibility with existing DER standards and demonstrates multimodal interoperability across heterogeneous protocols. Built on a fine-tuned LLaMA model trained on IEEE SEP 2.1, IEC 61968, and OpenADR, the IB performs seamless protocol conversion, message correction, and completion. We studied the effect of Retrieval-Augmented Generation (RAG) on the IB’s adaptability and compatibility. Integrating RAG reduced inference latency by about 500 ms, improving inference response time by over 90%. We showed RAG-augmented IB maintained high accuracy using 70% of the dataset and required no retraining when new payloads were introduced, lowering computational cost by 23% with only a small accuracy loss. The study concludes with future directions for reliable, real-time interoperability in distributed energy systems.