Estimation of Low-Frequency Oscillations Using Wide-Area Monitoring in Power System

Abstract

The presence of high gain excitation system causes Low Frequency Oscillations (LFOs) in the interconnected power system, which leads the system contingencies and operative with stressed conditions. Traditional, model base off-line approaches utilized Small Signal Stability Analysis (SSSA) to linearized the non-linear power system, which involve computational complexity as well as large computational time. The real time LFOs detection and mode estimation is not possible using SSSA. Moreover, in the conventional Supervisory Control Data Acquisition/ Energy Management System (SCADA/EMS), these methods utilize the result of the state estimators, which at present, are run at a periodicity of few minutes. Besides, the data scan rate of the SCADA system is of the order of 2-10 sec. The slow refreshing rate of the SCADA/EMS system limits its application in on-line stability analysis. To improve the power system mode estimation by providing the synchronized and GPS time-stamped phasor data with higher refreshing rate a synchrophasor based Wide Area Monitoring Systems (WAMSs) is being developed.
The methods available in the literature, for identification of low frequency critical modes under small disturbances, are broadly classified into the ambient data and ringdown or probing data based methods. The methods, which utilize ambient data, require a time window of about 10-20 minutes and are not much accurate at estimating the damping of the modes, whereas, the methods based on the ringdown or probing data, are more accurate and require small time window of about 8-20 sec for the estimation of the low frequency modes. Most of the work on the use of ringdown data have assumed that the low frequency oscillations can be modeled as summation of the complex exponentials. Prony method and its variants have been widely applied in the literature to estimate these modes, but these methods are sensitive towards the presence of the noise. Most of these methods have used linear regression approach to estimate the modes, whereas the actual estimation problem can be formulated as non-linear least square regression problem. Also the sensitivity of these methods towards the presence of colored Gaussian noise has not been considered in the literature. Since the power system also experiences large disturbances, such as fault or outage of a major generating plant, it is important to predict the transient stability and the stability margin of the system in real time. The Phasor Measurement Units (PMUs), forming a part of the synchrophasor based WAMS, can estimate generator rotor angle in real-time for predicting the angular stability of the power system. Repeated off-line simulations are performed to get the critical clearing time, considered as one of the measures of the transient stability margin. However, the time required by these methods is considerably high and also, sometimes, have the problem of numerical instability. The above limitations have resulted in the development of energy function based direct method of stability analysis.
Recently, neural network based methods have been suggested for the on-line stability prediction utilizing synchrophasor measurements, but these methods also face the problem of insufficient training due to non availability of enough real time data. Because of the certain limitations of the existing methods, as discussed above, in the design of the low-frequency mode estimator, the small-signal and the transient stability prediction, the main objectives behind the research work carried out to develop an estimators, which can accurately estimate the low-frequency modes of oscillation in the presence of Additive White Gaussian Noise (AWGN) and Additive Colored Gaussian Noise (ACGN). The real-time execution of the estimator can be achieved by using an adaptive filtering approach based on Least Mean Squares Sign-Data(LMSSD) algorithm. This approach not only helps to reduce the noise but also helps to retain the information within the permissible limit. In addition, a more reliable estimator based on signal subspace approach i.e.
Karhunen-Loeve Transform (KLT) is proposed, which not only provided the accuracy in estimation of the low frequency oscillating modes of the power system by completely decorrelate the signal from the noise also maximally compact the information contained in the signal. KLT defines a new axis of reference to present the signal subspace and further, this signal subspace is processed to obtain the low-frequency oscillating modes of the power system Estimation of Signal Parameter using Rotational invariance technique (ESPRIT).
To improve the performance of the above estimator for a signal having low SNRs i.e. 5dB and 0.5dB a Hample filter (HF) is utilized. The signal filtered by using HF is further processed using Modified Karhunen-Loeve Transform (MKLT) to achieve final variance by differentiating the dominant eigenvalue of the auto-covariance matrix with respect to final instant "N". To develop an estimator that can work in a real-time environment quite nicely with less computational complicity and can be applicable to a large power system in both ambient and ring down conditions.
The thesis emphasizes on developing various robust mode estimation techniques for a small and large interconnected power system at different real-time conditions such as short circuit fault and sudden load removable. Therefore, when there are outliers and high noise interruption, the KLT-TLS-ESPRIT is the best choice among all the proposed estimation algorithms mentioned in the thesis to achieve robust effective LFO modes estimation in power system.