Quantitative spectral analysis of dissolved gas in transformer oil based on the method of optimal directions

Autor: Gang Peng, Zhong Zhenxin, Songping Tang
Rok vydání: 2016
Předmět:
Zdroj: 2016 35th Chinese Control Conference (CCC).
Popis: The dissolved gas in transformer oil, which could represents the transformer faults, can be analysed by spectroscopy. Since spectral data with numerous wavelengths unavoidably contains useless and redundant information, the accuracy of spectral analysis would be affected and the model complexity would be increased. This paper proposes the quantitative spectral analysis of dissolved gas in transformer oil based on the method of optimal directions (MOD), in which the original spectral signals are processed by sparse representation, namely, they are represented by the sparse linear combinations of the atoms of MOD dictionary. Then, the prediction model of component concentrations is built based on the MOD dictionary, which is adaptive to the characteristics of the signals and contains all the key information of spectral data. Moreover, MOD dictionary is learned from the original spectral data by a two-step iterative optimization process with minimizing the sparse representation error under sparsity constraints. We use a dissolved gas spectral dataset of a real transformer oil in the experiments to evaluate this approach. Partial least squares (PLS) is performed on the MOD dictionary for component prediction. The experimental results verify that the proposed model can predict the component concentrations of the dissolved gas in transformer oil correctly and has high effectiveness.
Databáze: OpenAIRE