基于多级数据融合的核电站运行功能健康状态评估
Health Status Assessment on the Operation Function of Nuclear Power Plants Based on Multi-level Data Fusion
蒋翔宇 1冯毅雄 2洪兆溪 3胡炳涛 1司恒远 4谭建荣1
作者信息
- 1. 浙江大学流体动力基础件与机电系统全国重点实验室 杭州 310027
- 2. 浙江大学流体动力基础件与机电系统全国重点实验室 杭州 310027;贵州大学省部共建公共大数据国家重点实验室 贵阳 550025
- 3. 浙江大学流体动力基础件与机电系统全国重点实验室 杭州 310027;浙江大学宁波科创中心 宁波 315100
- 4. 深圳中广核工程设计有限公司 深圳 518172
- 折叠
摘要
随着新一代信息技术和人工智能技术与工业日益深度的融合协作,工业系统的运维从传统的人工定期运维向基于状态的智能维护迈进,其中健康状态评估作为智能运维的关键环节,通过源源不断的数据对系统的演化进行持续感知分析,为预测性维护提供决策支持.以核电站为研究对象,提出面向多部门协同环境下设备-系统-子功能-功能多层级融合的健康状态评估框架.针对工业系统控制的设备群具有大量状态参数的特点,提出基于偏差赋权的加权平均融合算子融合设备级参数,可快速定位并突出异常设备参数,避免评估结果因融合参数较多而展现"健康"假象;鉴于系统在不同健康状态下样本数据量的差异,提出考虑不平衡样本的非对称多分类学习方法融合系统健康值,多个健康等级之间通过信息有向迁移建立可自学习优化的健康评估模型,实现子功能健康状态及时自感知评估:最后基于集成学习融合多个子功能所属的健康评估器,得到宏观的运行功能健康状态评估决策.在提出的框架指导下对核电站反应性功能进行逐级健康状态评估,验证所提方法的有效性.
Abstract
With the new-generation information technologies and artificial intelligence that are deeply integrated with the industry,the maintenance of industrial systems is moving from the manual regular maintenance to state-based intelligent maintenance(IM).Health status assessment(HSA)is a key link in IM.Systemic states are recognized via the monitored data whereby provide decision support for maintenance.Taking nuclear power plants(NPP)as the research object,an HSA framework is proposed based on multi-level integration of equipment,system,sub-function and function in a multi-department collaboration.Due to equipment groups with numerous state parameters,a weighted average fusion operator based on deviation weighting is proposed to fuse equipment-level parameters,which can timely highlight the abnormal equipment.Considering the different amount of data in different health states,an asymmetric multi-class learning method under imbalanced datasets is proposed to fuse the systems'health values.The self-learning HSA model is established by the information transfer between multiple health levels,so that the health status of sub-functions can be self-perceived and assessed in a timely manner.The health assessors of multiple sub-functions are fused based on ensemble learning to obtain a macro-level operational function HSA decision.Exemplified with the reactivity function of the NPP,the proposed HAS frame is verified effective.
关键词
工业系统/健康状态评估/数据融合/核电站/基于状态的维护Key words
industrial system/health status assessment/data fusion/nuclear power plant/state-based maintenance引用本文复制引用
基金项目
浙江省"尖兵""领雁"研发攻关计划(2023X01214)
国家自然科学基金(52130501)
国家自然科学基金(52105281)
出版年
2024
NETL
NSTL
万方数据