Study On Harmonics Detection On Power System Of Railway Based On TSVM

With the electronic components and other non-linear devices or equipment is widely used in engineering practice, a large number of harmonic signal is injected into the public power system, making the power system signal distorted, endangering the safe and reliable operation of power system. Using a lot of electronic components and other non-linear devices or equipment, electrified railway has become one of the main sources of harmonic. In addition, due to the rapid development and using power in moving, compared to other harmonic sources, the electric power railway at aspects in generation quantity and aspect regions of harmonic is prominent. In common,harmonic detection methods of electric power railway come from ones of public power system, however, disadvantage always exists. For example, The Fourier transform harmonic detection method has spectrum leakage and fence effect, and low frequency resolution; The wavelet transform harmonic detection method has band aliasing, and the selection of mother wavelet is difficult; Harmonic detection method based on artificial intelligence, requires a lot of training samples, and there is the problem of local minima during training; Harmonic detection method based is affected by anti-noise easily and there is the problem of spurious frequency components.Considering the disadvantage of normal harmonic detection method and the feature which there are high frequency harmonic, a new harmonic detection method of electric power railway is established based on twins support vector combining with MSWF-ESPRIT. The method has high accuracy and high resolution with low computation, good anti-noise performance. The paper introduced twins support vector into harmonic detection by the first time, and to reduce the computation, TLS method is combined into TSVM. By simulation, the new TSVM is proved to have better accuracy and less computation. TSVM cannot calculate frequency parameters, and frequency is needed in TSVM, if we supposed frequency parameter, computation would be higher and accuracy would be worse, so ESPRIT which has high resolution is used to detect frequency. In addition, multi-stage wiener filter dimension reduction method is used before ESPRIT, and a new online decision prevention of filter stage is established, so the computation is lower. By simulation, the new method is proved to be effective and the influence condition and its pattern is shown. In the last, a new harmonic detection system of an experimental train is made, and through the electeddata, feature of every main loop of the train is shown.