A Power Equipment Maintenance Method Using Rule-Constrained Risk-Aware Preference Reinforcement Learning

LUO Haichen; LI Mingxuan; RONG Mingzhe; YANG Aijun; WANG Xiaohua

doi:null

您当前的位置：

首页 >

文章列表页 >

A Power Equipment Maintenance Method Using Rule-Constrained Risk-Aware Preference Reinforcement Learning

更新时间：2026-08-14

- A Power Equipment Maintenance Method Using Rule-Constrained Risk-Aware Preference Reinforcement Learning
- JOURNAL OF XI’AN JIAOTONG UNIVERSITY (2026)
- 作者机构：
  
  西安交通大学电工材料电气绝缘全国重点实验室，710049，西安
- 作者简介：
- 基金信息：
- DOI：
  CLC： TM 507
- Received：24 June 2026，
  
  Revised：2026-08-13，
  
  Accepted：14 August 2026，
- 稿件说明：
移动端阅览
LUO Haichen, LI Mingxuan, RONG Mingzhe, et al. A Power Equipment Maintenance Method Using Rule-Constrained Risk-Aware Preference Reinforcement Learning[J]. JOURNAL OF XI’AN JIAOTONG UNIVERSITY,2026.
DOI：

LUO Haichen, LI Mingxuan, RONG Mingzhe, et al. A Power Equipment Maintenance Method Using Rule-Constrained Risk-Aware Preference Reinforcement Learning[J]. JOURNAL OF XI’AN JIAOTONG UNIVERSITY,2026. DOI：

摘要

针对电力设备智能运维中多模态证据融合困难、通用大模型运维规程约束不足，以及高风险任务输出偏好难以控制等问题，提出一种面向电力设备运维的规程约束风险感知偏好强化学习方法。该方法将任务问题、现场多模态证据、规程约束及设备与场景元信息组成结构化运维输入，建立状态判断正确性、证据一致性、规程一致性、风险保守性和建议可执行性五维偏好裁决标准，并将综合偏好得分、安全评分差距、任务复杂度和证据充分性纳入动态间隔偏好优化，使模型输出更保守、可复核且符合运维流程。所构建的数据集覆盖6种任务类别，包括输电线路感知、隔离开关状态监测、电力现场安全监测、变电站监测、电力器件识别和发电厂综合监测。研究结果表明：所提方法在电力设备运维验证集上的选择题准确率达到87.89%，相较初始策略模型准确率提升了13.56个百分点，相较近端策略优化模型提升了5.54个百分点，从而验证了所提方法在离线电力设备运维场景验证中的有效性。该研究为提升多模态大模型在高风险电力设备运维场景中的规程遵循能力与安全输出能力提供了一种可行方案。

Abstract

To address difficulties in multimodal evidence fusion

inadequate enforcement of maintenance rules in general-purpose large models

and hard-to-control output preferences in high-risk power equipment maintenance

this paper proposes a rule-constrained risk-aware preference reinforcement learning method. The method organizes task queries

multimodal field evidence

maintenance rules

and equipment and scenario metadata into structured inputs. It establishes a five-dimensional preference criterion covering state-assessment correctness

evidence consistency

rule compliance

risk conservativeness

and recommendation actionability. Comprehensive preference scores

safety-score gaps

task complexity

and evidence sufficiency are incorporated into dynamic-margin preference optimization to encourage conservative

verifiable

and procedure-compliant outputs. The constructed dataset covers six task categories: transmission-line perception

disconnector-state monitoring

on-site safety monitoring at power facilities

substation monitoring

power-component recognition

and comprehensive power-plant monitoring. Experimental results show that the proposed method achieves a multiple-choice accuracy of 87.89% on the power equipment maintenance validation set

exceeding the initial policy model and the proximal policy optimization model by 13.56 and 5.54 percentage points

respectively. These results validate the method for offline power equipment maintenance scenarios. This study provides a feasible approach to improving rule compliance and safe outputs of multimodal large models in high-risk maintenance.

关键词

Keywords

references

唐文虎 , 牛哲文 , 赵柏宁 , 等 . 数据驱动的人工智能技术在电力设备状态分析中的研究与应用 [J ] . 高电压技术 , 2020 , 46 ( 9 ): 2985 - 2999 .

TANG W H , NIU Z W , ZHAO B N , et al . Research and application of data-driven artificial intelligence technology in condition analysis of power equipment [J ] . High Voltage Engineering , 2020 , 46 ( 9 ): 2985 - 2999 .

周俊煌 , 黄廷城 , 谢小瑜 , 等 . 视频图像智能识别技术在输变电系统中的应用研究综述 [J ] . 中国电力 , 2021 , 54 ( 01 ): 124 - 134+166 .

ZHOU J H , HUANG T C , XIE X Y , et al . Review on the application of video image intelligent recognition technology in power transmission and transformation systems [J ] . Electric Power , 2021 , 54 ( 1 ): 124 - 134, 166 .

段俊峰 , 李晨坤 , 姚文轩 , 等 . 基于人工智能赋能的新型电力系统关键技术综述 [J ] . 湖南电力 , 2024 , 44 ( 01 ): 1 - 10 .

DUAN J F , LI C K , YAO W X , et al . Review of key technologies for new power systems empowered by artificial intelligence [J ] . Hunan Electric Power , 2024 , 44 ( 1 ): 1 - 10 .

盛戈皞 , 钱勇 , 罗林根 , 等 . 面向新型电力系统的电力设备运行维护关键技术及其应用展望 [J ] . 高电压技术 , 2021 , 47 ( 09 ): 3072 - 3084 .

SHENG G H , QIAN Y , LUO L G , et al . Key technologies and application prospects of operation and maintenance of power equipment for new power systems [J ] . High Voltage Engineering , 2021 , 47 ( 9 ): 3072 - 3084 .

刘传洋 , 吴一全 . 基于红外图像的电力设备识别及发热故障诊断方法研究进展 [J ] . 中国电机工程学报 , 2025 , 45 ( 06 ): 2171 - 2196 .

LIU C Y , WU Y Q . Research progress of power equipment identification and overheating fault diagnosis based on infrared images [J ] . Proceedings of the Chinese Society for Electrical Engineering , 2025 , 45 ( 6 ): 2171 - 2196 .

贺明强 , 靳君 , 关新宇 , 等 . 基于计算机视觉的电力设备状态监测与故障诊断策略分析 [J ] . 集成电路应用 , 2024 , 41 ( 02 ): 224 - 225 .

HE M Q , JIN J , GUAN X Y , et al . Analysis of computer vision-based strategies for condition monitoring and fault diagnosis of power equipment [J ] . Integrated Circuit Applications , 2024 , 41 ( 2 ): 224 - 225 .

谢庆 , 张煊宇 , 王春鑫 , 等 . 新一代人工智能技术在输变电设备状态评估中的应用现状及展望 [J ] . 高压电器 , 2022 , 58 ( 11 ): 1 - 16 .

XIE Q , ZHANG X Y , WANG C X , et al . Application status and prospects of new-generation artificial intelligence technology in condition assessment of power transmission and transformation equipment [J ] . High Voltage Apparatus , 2022 , 58 ( 11 ): 1 - 16 .

蒲天骄 , 乔骥 , 韩笑 , 等 . 人工智能技术在电力设备运维检修中的研究及应用 [J ] . 高电压技术 , 2020 , 46 ( 02 ): 369 - 383 .

PU T J , QIAO J , HAN X , et al . Research and application of artificial intelligence technology in operation, maintenance and overhaul of power equipment [J ] . High Voltage Engineering , 2020 , 46 ( 2 ): 369 - 383 .

陈浈斐 , 章黄勇 , 马宏忠 , 等 . 基于深度学习的电力设备故障诊断方法研究综述 [J ] . 电气自动化 , 2022 , 44 ( 01 ): 1 - 2+6 .

CHEN Z F , ZHANG H Y , MA H Z , et al . Review of deep learning-based fault diagnosis methods for power equipment [J ] . Electrical Automation , 2022 , 44 ( 1 ): 1 - 2, 6 .

谢庆 , 王春鑫 , 李帆 , 等 . 知识及数据驱动的电力一次设备健康管理方法综述 [J ] . 高电压技术 , 2024 , 50 ( 02 ): 605 - 620 .

XIE Q , WANG C X , LI F , et al . Review of knowledge- and data-driven health management methods for primary power equipment [J ] . High Voltage Engineering , 2024 , 50 ( 2 ): 605 - 620 .

刘开培 , 李博强 , 秦亮 , 等 . 深度学习目标检测算法在架空输电线路绝缘子缺陷检测中的应用研究综述 [J/OL ] . 高电压技术： 1 - 12 [ 2023-03-09 ] . DOI: 10.13336/j.1003-6520.hve.20220273 http://dx.doi.org/10.13336/j.1003-6520.hve.20220273 .

LIU K P , LI B Q , QIN L , et al . Review on the application of deep learning-based object detection algorithms in defect detection of insulators on overhead transmission lines [J/OL ] . High Voltage Engineering : 1 - 12 [ 2023-03-09 ] . DOI: 10.13336/j.1003-6520.hve.20220273 http://dx.doi.org/10.13336/j.1003-6520.hve.20220273 .

陈富国 , 杨爱军 , 马慧珍 , 等 . 高压隔离开关状态智能感知系统设计与实现 [J ] . 自动化技术与应用 , 2021 , 40 ( 08 ): 131 - 135+152 .

CHEN F G , YANG A J , MA H Z , et al . Design and implementation of an intelligent perception system for the state of high-voltage disconnectors [J ] . Techniques of Automation and Applications , 2021 , 40 ( 8 ): 131 - 135, 152 .

Lu J , Clark C , Zellers R , et al . Unified-IO: A Unified Model for Vision, Language, and Multi-Modal Tasks [J ] . arXiv preprint arXiv: 2206.08916 , 2022 .

Zhu X , Zhu J , Li H , et al . Uni-perceiver: Pre-training unified architecture for generic perception for zero-shot and few-shot tasks [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 2022 : 16804 - 16815 .

Alayrac J B , Donahue J , Luc P , et al . Flamingo: a visual language model for few-shot learning [J ] . arXiv preprint arXiv: 2204.14198 , 2022 .

Wang P , Yang A , Men R , et al . Unifying architectures, tasks, and modalities through a simple sequence-to-sequence learning framework [J ] . arXiv preprint arXiv: 2202.03052 , 2022 .

Liu H , Li C , Wu Q , et al . Visual instruction tuning [C ] // Advances in Neural Information Processing Systems . 2023 .

Bai J , Bai S , Yang S , et al . Qwen-VL: a versatile vision-language model for understanding, localization, text reading, and beyond [J ] . arXiv preprint arXiv: 2308.12966 , 2023 .

Lin B , Ye Y , Zhu B , et al . Video-LLaVA: learning united visual representation by alignment before projection [C ] // Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing . 2024 .

Schulman J , Wolski F , Dhariwal P , et al . Proximal policy optimization algorithms [J ] . arXiv preprint arXiv: 1707.06347 , 2017 .

Rafailov R , Sharma A , Mitchell E , et al . Direct preference optimization: your language model is secretly a reward model [C ] // Advances in Neural Information Processing Systems . 2023 .

Wang J , Li M , Luo H , et al . Power-LLaVA: Large Language and Vision Assistant for Power Transmission Line Inspection [C ] // 2024 IEEE International Conference on Image Processing (ICIP) . Piscataway : IEEE , 2024 : 963 - 969 . DOI: 10.1109/ICIP51287.2024.10648271 http://dx.doi.org/10.1109/ICIP51287.2024.10648271 .

Vaswani A , Shazeer N , Parmar N , et al . Attention is all you need [J ] . Advances in neural information processing systems , 2017 , 30 .

Sennrich R , Haddow B , Birch A . Neural machine translation of rare words with subword units [J ] . arXiv preprint arXiv: 1508.07909 , 2015 .

Dosovitskiy A , Beyer L , Kolesnikov A , et al . An image is worth 16x16 words: Transformers for image recognition at scale [J ] . arXiv preprint arXiv: 2010.11929 , 2020 .

Bertasius G , Wang H , Torresani L . Is space-time attention all you need for video understanding? [C ] // ICML . 2021 , 2 ( 3 ): 4 .

Papineni K , Roukos S , Ward T , et al . Bleu: a method for automatic evaluation of machine translation [C ] // Proceedings of the 40th annual meeting of the Association for Computational Linguistics . 2002 : 311 - 318 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Binocular Vision-Based Synchronous Measurement Method for Rotational Errors

Construction and Hydrogen Production Characterization of Ammonium-Tolerant Photosynthetic Bacterial Mutants

AnIn-Situ Constructed ZnS/Graphene Electrochemical Sensor for Neurotransmitter Dopamine Detection

Progress in Visualization Experimental Studies on Heat Transfer of Supercritical Fluids

Construction and Parameter Identification of a Viscoelastic Constitutive Model for Power Transformer Insulation Spacers

Related Author

No data

Related Institution

No data

⁰