由于核反应堆发电的特殊性,核电厂对于生产安全的敏感度远胜于普通电厂。作为日常运维的重要环节,核电机组运行状态监测,对于核电厂的安全稳定运行具有重要意义。当前核电机组状态监测主要采用预设固定阈值报警结合人工监盘的方式,这种...由于核反应堆发电的特殊性,核电厂对于生产安全的敏感度远胜于普通电厂。作为日常运维的重要环节,核电机组运行状态监测,对于核电厂的安全稳定运行具有重要意义。当前核电机组状态监测主要采用预设固定阈值报警结合人工监盘的方式,这种方式无法发现低于报警阈值的异常状态,同时存在一定程度的漏报风险。核电运行数据作为高维海量时序数据,具有正常样本和异常样本分布不均衡以及数据缺乏标签的问题,这限制了有监督深度学习方法的使用。提出了一种基于变分自编码器(variational autoencoders,VAE)构建的无监督深度学习模型对真实运行数据进行异常检测,通过正常运行数据学习正常模式下数据在隐空间的分布,并基于异常数据无法被良好重构的原理,通过重构误差的大小来判别当前状态是否异常。实验以核电机组化学和容积控制系统(removal-chemical and volume control system,RCV)中的上充泵为例,使用真实运行数据结合插入异常的方式对模型进行了验证,并与经典机器学习方法进行了对比。实验结果表明基于变分自编码器的模型能够有效检测到核电真实数据中的异常数据片段及离群点,检测精确率和召回率均高于90%,检测性能相对孤立森林和支持向量机等经典机器学习算法具有优势,具备一定的实用价值和研究意义。展开更多
The complex geographical environment in China makes its gravity signals miscellaneous.This work gives a comprehensive representation and explanation in secular trend of gravity change in different regions,the key feat...The complex geographical environment in China makes its gravity signals miscellaneous.This work gives a comprehensive representation and explanation in secular trend of gravity change in different regions,the key features of which include positive trend in inner Tibet Plateau and South China and negative trend in North China plain and high mountain Asia（HMA）.We also present the patterns of amplitudes and phases of annual and semiannual change.The mechanism underlying the semiannual period is explicitly discussed.The displacement in three directions expressed in terms of geo-potential spherical coefficients and load Love numbers are given.A case study applied with these equations is presented.The results show that Global Positioning System（GPS） observations can be used to compare with Gravity Recovery and Climate Experiment（GRACE） derived displacement and the vertical direction has a signal-noise-ratio of about one order of magnitude larger than the horizontal directions.展开更多
文摘由于核反应堆发电的特殊性,核电厂对于生产安全的敏感度远胜于普通电厂。作为日常运维的重要环节,核电机组运行状态监测,对于核电厂的安全稳定运行具有重要意义。当前核电机组状态监测主要采用预设固定阈值报警结合人工监盘的方式,这种方式无法发现低于报警阈值的异常状态,同时存在一定程度的漏报风险。核电运行数据作为高维海量时序数据,具有正常样本和异常样本分布不均衡以及数据缺乏标签的问题,这限制了有监督深度学习方法的使用。提出了一种基于变分自编码器(variational autoencoders,VAE)构建的无监督深度学习模型对真实运行数据进行异常检测,通过正常运行数据学习正常模式下数据在隐空间的分布,并基于异常数据无法被良好重构的原理,通过重构误差的大小来判别当前状态是否异常。实验以核电机组化学和容积控制系统(removal-chemical and volume control system,RCV)中的上充泵为例,使用真实运行数据结合插入异常的方式对模型进行了验证,并与经典机器学习方法进行了对比。实验结果表明基于变分自编码器的模型能够有效检测到核电真实数据中的异常数据片段及离群点,检测精确率和召回率均高于90%,检测性能相对孤立森林和支持向量机等经典机器学习算法具有优势,具备一定的实用价值和研究意义。
基金supported financially by the National Natural Science Foundation of China(41174063,41331066 and41474059)the CAS/CAFEA International Partnership Program for Creative Research Teams(KZZD-EW-TZ-19)the SKLGED Foundation(2014-1-1-E)
文摘The complex geographical environment in China makes its gravity signals miscellaneous.This work gives a comprehensive representation and explanation in secular trend of gravity change in different regions,the key features of which include positive trend in inner Tibet Plateau and South China and negative trend in North China plain and high mountain Asia（HMA）.We also present the patterns of amplitudes and phases of annual and semiannual change.The mechanism underlying the semiannual period is explicitly discussed.The displacement in three directions expressed in terms of geo-potential spherical coefficients and load Love numbers are given.A case study applied with these equations is presented.The results show that Global Positioning System（GPS） observations can be used to compare with Gravity Recovery and Climate Experiment（GRACE） derived displacement and the vertical direction has a signal-noise-ratio of about one order of magnitude larger than the horizontal directions.
