Work

Domain Ontology-Anchored Experience Memory Mechanism for Power System AI Agent

Year: 2026

Type: conference-paper

Abstract: Power system AI agents need operational experience accumulation. Existing memory systems treat all information homogeneously, lacking domain semantics, cross-device transfer, and safety guarantees. This paper proposes a domain ontology-anchored mechanism: structured experience anchored to ontology nodes; ontology-relation-based transfer (model, type, topology); and safety-constrained credibility management. Experiments on a simulated 110 kV substation (100 experiences, 30 queries) achieve 97.9% recall when incorporating transfer (vs. TF-IDF 19.8%, embeddings 42.8%). Direct retrieval without transfer already reaches 97.1% recall; transfer mainly benefits new device models. Transfer reliability follows model > type > topology. Safety layer intercepts 80% of violations. Experience injection supplements device-specific details not covered by regulations.

Source: International Conference on Electronics and Communication, Network and Computer Technology (ECNCT)

DOI: 10.1109/ecnct70535.2026.11661425

Authors: Jia Si, Shanshan Huang, Wei Zhang, Xingfeng Mi

Institution: NARI Group (China)

Language: English

Topic: AI-based Problem Solving and Planning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: hybrid

OpenAlex ID: W7204240526

FWCI:

Cites: 5

Cited by:

Related to: 10