Research On Online Analysis Methods Of Low-Frequency Oscillations In Power Systems

As the policy of "west-to-east electricity transmission, north-south supply for each other, nation-wide power grid interconnection, the separation of power plant and supplement" implementing, the capacity of grid gradually enlarging, addition of the fast high magnification excitation system applying, further deteriorating of the system damping. Therefore, the low-frequency oscillations problems are easy caused by a small disturbance, seriously threaten the safe stable operation of the grid, which is an important bottleneck for power outgoing capacity.Low-frequency oscillation analysis is an important part of low-frequency oscillation study, as to the basis of carried out inhibition study to low-frequency oscillation. Eigenvalue analysis the basic methods for low-frequency oscillation study, but its reliability depends on the accuracy of mathematical model, so the online analysis methods which don't depend on system's mathematical model are more realistic for large interconnected system.Prony method and Hilbert-Huang Transform method (HHT method for short) are two main methods for online identification of the power system low-frequency oscillation at present. Prony method basing on the thinking of exponential function fitting can directly estimate the amplitude, frequency, attenuation factor, and initial phase of signal according to sampling value. HHT method is made up of Empirical Mode Decomposition (EMD for short) and hilbert transformat, its core part is EMD. EMD is a process that separates fluctuations or trends with different scales from a signal, producing a series of data with different characteristic scales, therefore, it can realize linearization and smoothing process for non-linear non-stationary signal, getting to the Intrinsic Mode Function(IMF for short) component reflecting the original signal's local characteristics. The paper studies Prony method and HHT method, their advantages and disadvantages are comparably analyzed through simulation examples. The Prony method has perfect theory, clear physical significance, accurate results, but it is sensitive to noise. The HHT method with adaptive filtering properties, do not need to go straight, filtering and other data preprocessing, but it has end effection, modes confusion and other problems.For sampling data with noise, the paper presents an online identification method combining the EMD and Prony to analyze the low-frequency oscillations. First,decompose the complex signal with noise, get to IMF components with single scale; then, judge and separate noise components from these IMF components, do data reconstruction;last, analyze the reconstructing signal using Prony method, realize online low-frequency oscillation identification for sampling data with noise. Simulation results show that the comprehensive method is feasible and accurate, and has practical significance to the low-frequency oscillation online analysis.