论文部分内容阅读
北美电网监测系统（FNET）是在配网侧实时采集电网频率的广域测量系统。由于硬件故障或网络中断,频率扰动记录单元（FDRs）采集的数据不可避免地包含尖峰或缺失数据段等异常数据,在剔除尖峰同时常用的一维中值滤波,弱化频率波动的细节信息无法弥补缺失数据段。针对此一问题,提出了融合稳健统计和B样条函数的频率异常数据处理方法,它通过设定阈值辨识尖峰值,采用B样条基函数的线性组合重构原始频率序列,引入曲线粗糙度控制B样条基函数学习过程中存在的过拟合问题。该方法仅在局部范围内处理频率异常数据,能最大限度地保留频率波动信息,且计算简洁,能实现任意阶B样条函数的构造及学习,易于推广到其他时间序列的数据预处理。“,”Raw data from the frequency disturbance recorders （FDRs） inevitably contain outliers such as spikes or missing segments due to hardware failure and/or network interruption, and one dimension median filtering is commonly used to eliminate these outliers. However, this filtering removes the spikes as well as the detailed information of frequency variation, and is incapable of replacing the missing data. Consequently, we combined robust statistics and B spline function to deal with outliers in the FDR raw data. The spikes were first identified by a preset threshold of robust statistics, then the spikes and the missing data were replaced using B-spline smoothing, finally, the FDR data were reconstructed by a linear combination of a family of B spline basks functions; and roughness of the constructed curves was controlled to avoid the over fitting problem of this technique. The proposed method can only be used for frequency outliers and keep the rest of the FDR data intact. Test examples validate that the method and its application may be easily extended to time series data in other fields using arbitrary order of Bspline functions.