| Under the new era environment,the power system has become a wide area and large interconnected network.The long-distance transmission and the use of various rapid excitation equipment,as well as the immature control technology in the process of new energy access,have caused a series of phenomena that endanger the power system as low-frequency oscillations.Analysis of the mechanism of low-frequency oscillation is of great significance for maintaining the safe and sTable.operation of the power grid and its long-term development.This paper identifies and analyzes low frequency oscillations mode parameters base on the measurement data.Compared with the model-based method,the signal processing method base on the actual operation data thus reflect the real state of the grid.The calculation is simple and efficient,avoiding the difficulties of determination and calculation of the model when the power grid become large,provides important basis for real-time monitoring and control of the power grid.With the deepening of the signal analysis algorithms,a series of low-frequency oscillation mode identification methods based on measurement data have been developed.There are two main parts of analysis.One is the signal preprocessing,the other is identification.The signal preprocessing algorithm reduce the noise,or decomposes the signal and selects the valid signal.First,this paper compare the low-pass filter,wavelet denoising,EMD method,VMD method and generalized morphological filtering with the Signal-to-Noise Ratio(SNR)as a reference.Results show that the digital low-noise filter has the highest signal-to-noise ratio between the signal pre-processing methods above.Cause the low-frequency oscillation signals has a clear frequency distribution range,that is 0.1-2.5 Hz,so that the digital low-pass filter has the most effective effect.The use of sym8 wavelet-based wavelets for fixed-threshold noise reduction can significantly smooth the filtered signal,but the signal-to-noise ratio is not as good as direct low-pass noise reduction.Soft threshold denoising with sym8 wavelet-based wavelets alone can significantly smooth the filtered signal,but the signal-to-noise ratio is not as good as digital low-pass filter.Generalized morphological filtering with emi-circular structure and sinusoidal structure can effectively reduce Gaussian white noise,but the curve is smoothness.The EMD decomposition loses the energy of the effective signal and occur modal aliasing while filtering.The VMD decomposition become invalid when decomposes the small frequency interval and overlapping bandwidths signals.This paper selects digital low-pass filter and wavelet soft threshold denoising for signal preprocessing.Currently,there has been a series of application of matrix pencil method in low-frequency oscillation mode identification,and the main problem is modal order determination.This paper introduces currently available mode order identification method and analyze their effectiveness,summarize the singular value's alignment characteristics then propose geometric-ordered matrix pencil method based on inflectional angle's cosine value of odd singular value curve,results of tests show that the new method is more robust and intuitive.For further research,this paper compares the matrix pencil method with the SSI method.The results show that the matrix pencil method is simple and straightforward with single-channel data,but the SSI method is not accurate.The reason is that the SSI method corresponds to multi-channel data,so the matrix beam algorithm is better in the case of single-channel data.Finally,this paper takes four-machine and two-region models as examples,and combines the measured low-frequency oscillation data to analyze.Results show that the proposed algorithm is clear and efficient,and for low-frequency oscillation signals,directly use low-pass filters and wavelets denoising then use of the matrix pencil method can achieve good identification results. |
PDF Full Text Request