Application of Neural Network in Atmospheric Correction of High Voltage Test

During the high voltage test of external insulation, when the ambient temperature and humidity are relative large, the ratio of humidity to air density (h/δ) exceeds the given range by the IEC standard 60060-1, it's not feasible to perform atmospheric correction according to the method specified by the standard. This paper proposes a neural network model for atmospheric correction. First, a test platform was set to measure the propagation velocity of the streamer when a spark discharge occurs in the air under different atmospheric conditions and different impulse voltage, the velocity as one of the input features of neural network; then combined the experimental data from our laboratory to train the neural network; Finally the neural network model can be used to predict the atmospheric correction factor when the ambient temperature and relative humidity are large.