次同步振荡在线监测的同步提取变换和朴素贝叶斯方法

目前基于相量测量单元(PMU)实现次同步振荡在线辨识和告警存在的问题有:参数辨识一般只辨识频率、幅值,不辨识衰减因子;告警阈值的确定需要人为经验,导致告警判据的快速性和可靠性难以保证。针对上述问题,提出将同步提取变换(SET)和机器学习方法——朴素贝叶斯(NB)方法相结合的次同步振荡在线监测方法。SET可以快速、准确地辨识出次同步振荡的模态参数,而NB方法可以自动实现次同步振荡在线预警。首先,通过SET对已有的历史次同步振荡数据进行辨识,将辨识得到的频率和衰减因子交由NB方法学习,并生成NB分类器。然后,当有新的PMU上传的振荡信号数据时,先采用SET辨识出振荡的频率和衰减因子,再将这些参数交由NB分类器来判断是否发生次同步振荡,并准确预警,从而实现对次同步振荡的在线监测。通过IEEE第二标准模型验证了所提方法的有效性。

On-line Monitoring of Subsynchronous Oscillation Based on Synchroextracting Transform and Naive Bayes Method

At present, there are some problems for online identification and alarm method of subsynchronous oscillation (SSO) based on phasor measurement unit (PMU). Parameter identification method only identifies the frequency and amplitude, and does not identify attenuation factor. Alarm threshold values need to be determined by personal experiences. Therefore, it is difficult to guarantee the rapidity and reliability of alarm criterion. In view of the problems mentioned above, the on-line monitoring SSO method is proposed combing synchroextracting transform (SET) and naive Bayes (NB), which is one of the machine learning method. SET can quickly and accurately identify the mode parameters of SSO, and NB can automatically realize online warning. Firstly, SET identifies the historical data of SSO, and then NB method learns from the identified frequency and attenuation factor, which helps to generate a NB classifier. Secondly, when there are new upload oscillation data of PMU, the oscillation frequency and attenuation factor are identified by SET at the first step. Then, NB classifier is used to deal with these parameters to determine whether SSO occurs or not and realize accurate on-line early warning, so as to realize the on-line monitoring of SSO. The simulation of the second IEEE benchmark model verifies the effectiveness of the proposed method.