A fast current zeroes estimation algorithm for controlled fault interruption based on an improved BP neural network

Predicting zeroes precisely and rapidly after a fault initiation is the basis of controlled fault interruption. However, none available algorithms could predict current zeroes within several milliseconds. The objective of this paper is to propose a fast estimation algorithm that can predict current zeros within 3ms after fault initiation. An algorithm is proposed based on an improved BP network. In the proposed algorithm, initial phase angle of the fault was acquired by a wavelet transform. Both the angle and waveform of the fault current was set as an input of the neural network. The first current zero was set as an output. The neural network was trained by over 10,000 trails by using waveforms of fault current with different parameters acquired from short fault circuit simulations. Simulation results show that the first current zero is estimated within 3 ms after the fault initiation with a current prediction error of ±0.5 ms.