采用滑动窗口及多重加噪比堆栈降噪自编码的风电机组状态异常检测方法  

A Method for Detecting Anomaly Conditions of Wind Turbines Using Stacked Denoising Autoencoders with Sliding Window and Multiple Noise Ratios

作　　者：陈俊生[1] 李剑[1] 陈伟根[1] 孙鹏 Chen Junsheng;Li Jian;Chen Weigen;Sun Peng(State Key Laboratory of Power Transmission Equipment&System Security and New Technology Chongqing University,Chongqing 400044,China;State Grid Henan Electrical Power Research Institute,Zhengzhou 450000,China)

机构地区：[1]输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆400044 [2]国网河南省电力公司电力科学研究院,郑州450000

出　　处：《电工技术学报》2020年第2期346-358,共13页Transactions of China Electrotechnical Society

基　　金：国家重点基础研究发展计划（973计划）（2012CB215205）;高等学校学科创新引智计划（B08036）资助项目

摘　　要：该文提出一种基于多元变量数据重构的风电机组状态异常检测方法。针对风电场数据采集与监控(SCADA)系统数据,首先,建立基于滑动窗口的堆栈降噪自编码(SDAE)模型,在获取机组正常运行状态下变量间的互相关性和各变量短时相依性的基础上重构机组状态数据;其次,为提高模型特征学习能力,提出多重加噪比的SDAE模型训练方法学习机组状态参数的全局和局部特征;最后,采用重构误差的马氏距离为机组状态监测指标,通过核密度估计方法分析机组正常数据监测指标的概率密度分布,确定机组正常运行状态下监测指标的阈值,定义监测指标连续越限数监测机组状态,计算各状态参数对监测指标越限的贡献度,实现机组参数异常检测。华东某风电场SCADA数据分析结果表明该方法可有效地用于实际风电机组运行状态的异常检测。This paper presents a multivariable reconstruction approach to detect the anomaly conditions of wind turbines(WTs),focusing on the data collected from the wind farm supervisory control and data acquisition(SCADA)system.Firstly,the stacked denoising autoencoders(SDAE)model with sliding window was developed to capture nonlinear correlations among multiple variables and temporal dependencies at each variable simultaneously,and reconstruct the status data of WTs.Secondly,the SDAE model with sliding window was trained under multiple noise ratios to learn the coarse-grained and fine-grained features of condition parameters for improving the feature representation of model.Finally,the Mahalanobis distance(MD)of reconstruction error was defined as the monitoring index.The threshold of monitoring index was obtained based on the probability density function(PDF)of MD with the kernel density estimation.The duration of over-limit was utilized to detect the anomaly conditions of WTs.The anomaly conditions of WTs were identified based on the contribution of each variable.The analysis results of SCADA data collected from a wind farm in Eastern China show that the proposed method is effective for the anomaly condition detection of actual wind turbines.

关 键 词：风电机组 异常检测 数据采集与监控系统 堆栈降噪自编码 滑动窗口 多重加噪比 

分 类 号：TM315[电气工程—电机]