An outlier robust detection method for online monitoring data of dissolved gases in transformer oils

The concentration of gas in insulation oil is the main basis for judging the health status of transformers. A robust detection method for outliers based on online monitoring data of dissolved gas in transformer oil is proposed to judge the operation status of transformers more accurately. The online detection device for dissolved gas in transformer oil based on tunable laser absorption spectroscopy (TDLAS) multicomponent gas is designed. Based on the near-infrared absorption band of the basic fault characteristic gas of the transformer, lasers with different wavelengths are selected, combined with semiconductor sensors to measure the characteristic gas, and the second harmonic wave to the gas phase concentration is completed based on the piecewise linear fitting of the least squares, the conversion calculation from gas phase to liquid phase enables online monitoring data acquisition of dissolved gas in transformer oil. Based on the robust statistical theory and the characteristics of the abnormal value of the online monitoring data of dissolved gas in transformer oil, a robust multivariate detection method of minimum covariance determinant (MCD) for the abnormal value of characteristic gas is proposed. This method uses the idea of iteration and Mahalanobis distance to construct a robust covariance estimator, detect the abnormal value, and classify the abnormal and normal data. To solve the problem that the MCD algorithm may fain under specific conditions in the detection process, an optimization algorithm of MCD algorithm - high-dimensional robust covariance matrix robust estimation algorithm (MRCD) is proposed to improve the robustness of the detection process. The experimental results show that the absolute error of the on-line monitoring data acquisition results of dissolved gas in transformer oil is controlled within 2 mu L/L, and the abnormal values in the monitoring data can be accurately detected, and the detection reliability is high, which meets the requirements of practical application.