Noise suppression for weak current measurement based on neural-network-assisted UHV FOCS

Abstract

Weak current measurement plays an essential role in the power grid systems, which remains challenging to be achieved with fiber optic current sensor (FOCS) due to ambient and internal noise. In this work, noise suppression method based on back propagation neural network (BPNN) is proposed to calibrate FOCS which is originally applied to ultra-high voltage (UHV) systems so that output accuracy can be improved when it is utilized to detect weak current. The output errors induced by noise, the feature of FOCS output signal and the reason for choosing BPNN are deeply analyzed. Performance of BPNN with different parameters is also investigated and results show that BPNN with 2 hidden layers and 3 neurons meets the requirements of high efficiency and high accuracy in our application. Then, 15750 groups of FOCS output data are collected under different temperature and utilized for network training. With the well-trained BPNN, weak current as low as 0.1 A is successfully detected by UHV FOCS when temperature varies from -30 °C to 70 °C, and ratio error (RE) is limited between -0.2% and 0.2%, showing a better performance on noise elimination than traditional Gaussian filter and Fourier filter.