Application and performance evaluation of multimodal large model agents in power equipment operation and maintenance scenarios

Read full text Summarise with VIRAT
Authors: Wenzheng Song, Xiya Li, Peng Gao
Cited by: 0
References: 7

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract This paper aims to address issues in the application of general multimodal large models in power equipment operation and maintenance scenarios, such as insufficient professionalism, inadequate task planning capabilities, and weak safety risk awareness. A multimodal large model agent, PowerAgent, tailored for power operation and maintenance, is proposed. This approach is based on the LLaVA-Transformer architecture, enhancing professional knowledge through a visual-text feature alignment mechanism with the integration of a power knowledge graph. The ReAct paradigm is adopted to construct a hierarchical task planning and tool invocation framework to ensure the execution of complex tasks. LoRA technology is used for domain adaptation and fine-tuning of safety protocols to ensure output safety. Experimental results show that PowerAgent achieves a state recognition F1 score of 0.92, a task chain completion accuracy of 80%, and a safety compliance rate of 100%, significantly outperforming the baseline model Qwen-VL-Chat. It also demonstrates near-human expert-level comprehensive analysis capabilities in typical cases, such as transformer bushing oil level analysis and circuit breaker mechanical characteristics diagnosis. This research validates the effectiveness and practicality of multimodal large models in power operation and maintenance scenarios and provides a reliable technological pathway for the intelligent transformation of the industry.

Fields of Study & Subjects

Multimodal Machine Learning Applications Power Line Inspection Robots Advanced Neural Network Applications

Publication details

Journal / source
Journal of Physics Conference Series
Volume
3192
Issue
1
Pages
012030–012030
Publication date
1 March 2026
Language
EN