基于FCM聚类的随机子空间低频振荡模态识别算法

Recognition Algorithm for Low-frequency Oscillation Mode Through Stochastic Subspace Based on FCM Clustering

振荡模态的精准捕捉对有效抑制低频振荡有重要意义,基于量测的低频振荡模态辨识方法在在线监测识别领域具有广阔应用前景。本文针对模态识别算法定阶困难、易存在虚假模态等问题,提出了基于模糊C均值聚类的多阶随机子空间算法。通过多阶子空间计算可捕捉所有可能的系统模态,并通过模糊C均值算法确定实际最低阶数,经虚假模态筛除确定最终振荡主导模态,并且能降低干扰,提升辨识抗噪性能。本文算法与Prony算法进行了性能对比,并通过四机两区系统和实际电网相量测量单元量测数据验证了算法的适用性和鲁棒性。

The accurate capture of oscillation modes is of great significance to effectively suppressing the low-frequency oscillations,and the low-frequency oscillation mode recognition method based on measurement has a broad prospect in the field of online monitoring and identification. In this paper,a multi-order stochastic subspace algorithm based on fuzzy C-means(FCM)clustering is proposed to solve problems such as the difficulty in order determination using the modal recognition algorithm and the existence of false modes. This method can capture all possible system modes through multi-order subspace calculations,determine the actual minimum order using the FCM algorithm,and determine the final dominant oscillation mode through false mode screening. In addition,it can reduce the interference and improve the anti-noise recognition performance. The comparison of performance between the proposed algorithm and the Prony algorithm is conducted,and the applicability and robustness of the novel algorithm are tested by simulations of a four-machine two-area system and the actual phasor measurement unit(PMU)measurement data of power grid.