Investigation on mathematic model of corona current based on the curve fitting by artificial neural network

In order to study the general mathematic model for corona currents, we designed a back propagation artificial neural network (BPNN) consisting of two layers which can approximates to an arbitrary function with an arbitrary accuracy to fit the measured corona currents. Theses current waveforms are represented by double exponential function, Gaussian function, and random irregular pulses. The results indicate that the BPNN can fit the experiment corona current waveforms with a high accuracy when the neuron number is selected from 5 to 10. Compared with the measured current waveforms, the error of mean square (MSE) of the fitting current waveforms can arrive 10-4 and the calculation time is about 2 to 10 seconds by the BPNN method. The analytical expressions for the corona currents can be achieved via extracting the weights and thresholds parameters of the BPNN. The expressions can be used as the general mathematic model for corona currents because the expressions have the same structure and the structure is independent of the waveform. ©, 2015, Gaodianya Jishu/High Voltage Engineering. All right reserved.