Application And Research On Theory Of Lifting Wavelet In Noise Elimination And Pattern Recognition Of Partial Discharge Signal

The partial discharge (PD) inside insulation of the electrical equipment is the important factor leads to the electrical equipment insulated deterioration. The disfigurement styles of the insulation inside electrical equipment close to the discharge pattern. PD monitoring on-line and patten recognition technology are a key way to discover the insulation faults caused by PD in the electrical equipment and to diagnose the inside disfigurement in the insulation system of electrical equipment timely. Its important significances are that the insulation accidents of high voltage electrical equipment can be defended and its safety and reliability can be improved. The bottleneck problems of the PD monitoring on-line in the electrical equipments are that how to correct extract signal availably and to recognise discharge pattens of PD. Many experts have presented a lot of methods for PD signal denoising and patten recognition, there are no availability method for the actual projects. This paper shows the some ways of PD signal denoising and patten recognition, introduced the basic theory of somemethods of signal denoising with the wavelet transform——wavelet analysis. Andscope of application of some typical remove noise processing methods are analyzed, compared, discussed. The generally, flexibility and highly effective lifting wavelet is introduced to the PD signal denoising and the patten recognition, the characteristic of lifting wavelet is analysed. The constructes steps and methods of the tradtion wavelet with lifting scheme, the adaptability cross lifting wavelet transform theory with the lifting schem is researched. The lifting method is putted into the detection of signal denosing and singular point. Simulation results shows that the method has the definitepracticability. A new way of noise elimination——adaptability cross lifting wavelettransform theory is presented in this paper. This method first predicts the odd coefficients from the even coefficients, as per the standard lifting construction; then predicts the even coefficients from the odd coefficients. This is accomplished by shifting the input sequence by one and then feeding it into the same lifted transform, simulation experiments and scene datas shows that it has a definite practical prospect. Add, this paper give a way that decomposite the single time domain impulse signal with lifting wavelet, 8 characteristic quantities including: the top three serial number of energy and entropy, respective researches are brought forward as input of neural network, 3 kind of PD signals's pattern recognition results with BP neural network are studied. The recognition via simulation experiments to prove that the neural network recognition methods with these characteristic quantity are simply, effective and practical. It can give a reference for the PD signal recognition.