Abstract:

Aiming at the problem that the calculation speed of the singular value decomposition denoising algorithm for constructing Hankel matrix is slow, an improved singular value decomposition denoising algorithm is proposed. The algorithm increases the delay step of data points between rows of Hankel matrix to reduce the attractor track matrix andaccelerate the calculation speed. Besides, the effective order of singular values is obtained by combining the difference spectrum method of singular value energy and the mean value method of singular value energy to ensure good denoising effect. On this basis, a series arc fault detection method of photovoltaic system based on improved singular value decomposition denoising algorithm is proposed. Firstly, the improved singular value decomposition denoising algorithm is used to remove the noise. Secondly, the FFT and wavelet transform are used to analyze the signal. The sum of harmonic amplitude values and the energy of wavelet detail coefficients d₂ and d₃ are selected as the feature vectors. Finally, the linear support vector machine (LSVM) is used to realize arc detection. The experimental results show that the propsed method has high accuracy.

Key words: photovoltaic system, arc fault, improved singular value decomposition, feature vector, support vector machine (SVM)