Online Identification of Low-Frequency Oscillations Based on Improved Multi-Signal Prony Algorithm

An improved multi-signal Prony algorithm suitable to on-line identification of low frequency oscillation is proposed. Firstly, by means of wavelet transform the noise contained in each signal is eliminated; then the DC component is also eliminated and sample function matrix for multi-signal is built up, the Prony algorithm is improved by means of singular value decomposed-total least square (SVD-TLS) and the signal space is separated from noise sub-space, and then the order number of signal is determined; finally, the identification is carried out by least square. The ideal signal, simulated signal and practical recorded signal are analyzed by traditional Prony algorithm, the improved single signal Prony algorithm and the improved multi-signal Prony algorithm respectively, analysis results show that the identification accuracy can be improved while signals are simultaneously analyzed by the proposed improved multi-signal Prony algorithm, the calculation time is reduced, and the identified order number as well as the identified result are in advance of traditional algorithms, so the proposed algorithm is suitable to on-line identification of low frequency oscillation.