Rule Verification Method for Power Operation Order Instructions based on Large Language Model

In order to identify the temporary additional rules that may appear for specific operation tasks to prevent errors in substation operation instructions, a rule verification method for power operation order instructions is established based on a large language model (LLM). First, the paper uses the graph isomorphism algorithm to form an error-prevention knowledge graph (EPKG) that integrates the physical structure information and rule information of the power grid, so as to improve the analysis ability of the LLM on the power grid topology; the paper also uses retrieval-augmented generation (RAG) technology, which is used to improve the understanding ability of the LLM on the operation rules, to build a vector database for the temporary additional rules of the operation task. Then, by extracting the argument information of operation instructions and using the argument post-processing method, the operation instruction graph is obtained which can be located on the EPKG to realize the graph-based instruction verification. Furthermore, the corresponding additional rules related to an instruction that can be understood by LLM are extracted from the RAG knowledge base to realize RAG-based instruction verification. Finally, in the case analysis, the operation of a certain device operation instruction is simulated and deduced, and interpretability of the graph-RAG-based rule verification method is demonstrated.