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。
The North American Power Grid Monitoring System (FNET) is a wide area measurement system that collects the grid frequency in real time on the distribution network side. Due to hardware failure or network interruption, the data collected by Frequency Disturbance Recorded Units (FDRs) inevitably contain abnormal data such as spikes or missing data segments. The one-dimensional median filter, which is commonly used at the same time as eliminating spikes, can not make up for the details of weakening frequency fluctuations Missing data segment. In order to solve this problem, we proposed a frequency anomaly data processing method based on the fusion of robust statistics and B-spline functions. By setting thresholds to identify spikes and reconstructing the original frequency sequence by linear combination of B-spline basis functions, Control Over-fitting Problems in B-Spline Basis Function Learning Process. This method only deals with the frequency anomaly data in the local scope, preserves the frequency fluctuation information to the maximum extent, and the calculation is concise. It can construct and learn the B-spline function of any order and is easy to generalize to the data preprocessing in other time series.