Study On Modal Identification For Power System Low-frequency Oscillations Based On Wide-area Stochastic Responses

The interconnection of power grids forms a typical system of ultra-high voltage, long-distance, large-capacity and AC/DC hybrid operation. The network topology becomes more complex and the operation of power system is quite flexible, resulting in low-freqyuency oscillation events occur frequently. The system stochastic responses are excited by samll random fluctuations in loads and other related small variations when the system operates in normal conditions. Since the stochastic responses are continul avarialbe and are noninvasively measured, the abundant data can timely characterise the current dynamic properties of the power system. Therefore, this study is foucs on the modal identification of low-frequency oscillations using the system wide-area stochastic responses, in order to mine and utilize effectively the wide-area measured spatio-temporal data in large power grids, as well as to track the operating state of the system. The mian work in the dissertation is as following:(1) The mode paremeters estimated from the system stochastic responses are assumed to be a set of normally distributed stochastic variables. To solve the problem in most mode estimation methods that only give the point estimetes of mode pameters, an interval estimation method for the oscillation mode parameters is presented using single channel stochastic response. The auto-regressive and moving average (ARMA) model of the single channel stochastic response is identified by using the prediction error method (PEM), and on this basis the relation between the covariance matrix of auto-regressive (AR) model parameters and the variance of mode parameter is derived according to the estimation procedure of the mode parameters from the AR parameters. Finally, the mode parameters and their variances for each modal are calculated, then the confidence intervals of the mode frequency and damping ratio are constructed.(2) As the single channel stochatic responses are not of good observability for multiple system oscillation modals, an interval estimation method for the oscillation mode parameters is presented using multiple channel stochastic responses. The mode parameters are obtained by covariance-driven stochastic subspace identification (SSI-COV), and the covariances of the Toeplitiz matrix are derived from the multiple channel measured signals. Then base on the first order perturbation method, the covariances of the Toeplitiz matrix are progatate to the covariances of the observability matrix, the system state matrix and the mode parameters.Finally, the confidence intervals of the mode paramters for multiple oscillation modals are estimated from single set of multiple signals.(3) There are two interesting phenomena widely reported in the mode estimation using system stochastic responses: (i) the frequency estiamtes are more accurate than the damping ratio estiamtes; (ii) the accuracy of the estimates improves under the lighter damping conditions. In order to to investigate the estimation accuracy of mode estimation, Monte Carlo simulations in the two-area system are carried out and the influences of varying the mode parameters on the estimation accuracy are compared and analyzed from perspective of variance propagation. In addition, theoretical explanations for the observations in ambient mode estimation are also provided. The investigations and discussions of the estimation accuracy reveal the underlying characteristics in the mode estimation under ambient conditions.(4) Mode shapes are important parameters to characterise the low-frequency oscillation modals. In order to fully use the correlative characteristics of the wide-area spatio-temporal data, the relationship between vector autoregressive (VAR) model of wide-area multiple channel stochastic responses and system oscillation modals was discussed, and an oscillation mode shape estimation method is presented based on the VAR model. The vector autoregressive model parameters are estimated by least square method via the implementation of QR decomposition; and then the oscillation mode shapes are calculated;finally, the dominant modals are determined according to the peaks of power spectrum of the system stochastic responses. As the channels of the system output gradually increase, the computation of the mode shape estimation becomes more complex. Therefore a new mode shape estimation method is further studied base on empirical orthogonal function analysis,improving the estimation efficiency.(5) On the basis of the theoretical research for the modal identification of low-frequency oscillation, a software titled "WAMS based low-frequency oscillation analysis and warning software" is developed. In this software,the oscillation modals are identified using systen stochatic responses, and early warning is provided when the a certain modal has a small damping ratio. The software provide important information for the system operators to take appropriate control measures, to improve the damping of the associated modal and to prevent the out-of-step oscillation accident. Besides, the softeware also other modules such as oscillation real-time monitoring, offline transient event analysis, statistical reports, et al.The modual functions in the software meet the requirements, which are verified by the Experimental Verification Center in Electric Power Research Institute of State Grid Corporation.The theoretical research and software development in this dissertation are of great theoretical significance and application value for enriching the low-frequency oscillation analysis theory, for enhancing the inter-area transmission capacity and for avoiding blackouts of the large power systems.