高电压技术2026，Vol.52Issue(7)：2986-2997,12.DOI:10.13336/j.1003-6520.hve.20260632
物理-知识双约束下的电力具身智能作业决策技术
Electric Power Embodied Intelligent Operation Decision-making Technology Under Dual Constraints of Physics and Knowledge
摘要
Abstract
This paper proposes a physics-knowledge dual-constraint decision-making technology for power-embodied intelligent operations,targeting the prominent problems in practical scenarios.Sudden state changes of equipment during power distribution cabinet sequential control easily cause skip-level operations,and trajectory deviation in UAV line in-spection may lead to insufficient safety distance.In addition,existing methods mostly implement safety specification verification in a post-event manner,which fails to realize full-process proactive safety prevention.The proposed method transforms spatial geometry,mechanical balance,and electrical safety requirements into computable physical consistency constraints,and converts operation timing,compliance criteria,and equipment limits into rigid safety specification con-straints.A closed-loop decision loop consisting of constraint verification,disturbance classification,local replanning,action pre-verification,and state confirmation is constructed.Combined with dynamic constraint prompt engineering,adaptive action correction is achieved without model retraining.Verified by Isaac Sim,AirSim,and field experiments,the method obtains a correct response rate of 94.5％under 200 random disturbances in cabinet control scenarios and achieves a task completion rate of 98.11％and an image availability rate of 92％in 53 UAV field tests,without point cloud scan-ning or teaching flights.The proposed method pre-embeds physical feasibility and operational compliance into action generation and execution,providing verifiable safety decision support for power-embodied intelligent operations.关键词
电力具身智能/物理一致性约束/业务安规约束/配电柜顺控/无人机巡检Key words
power embodied intelligence/physical-consistency constraint/operational-safety constraint/distribution cabinet sequential control/UAV inspection引用本文复制引用
谈元鹏,宋睿,张鹏,莫文昊,王征..物理-知识双约束下的电力具身智能作业决策技术[J].高电压技术,2026,52(7):2986-2997,12.基金项目
国家电网有限公司科技项目(520900260043-237-ZN).Project supported by Science and Technology Project of SGCC(520900260043-237-ZN). （520900260043-237-ZN）
