Abstract: The fixed residual current effective value is used by the residual current electrical fire monitoring system as the early warning criterion of electrical fire, and there are great safety risks because of frequent false alarms in the process of using it. In this paper we built an electrical fire experiment platform to extract the residual current waveforms of normal operation and earth fault under different loads, and proposed a method by using the characteristic quantity of residual current waveforms to warn electrical fire. Wavelet transform is used to filter the original residual current signal and extract the low-frequency component. The fault feature vector is formed from the time-domain characteristics and waveform characteristics of the low-frequency component, and it's used to train BP neural network. Through intelligent algorithms automatically identify faults, the accuracy of residual current fire warning is improved.

Key words: electrical fire monitoring, db20 wavelet transform, BP neural network, residual current waveform characteristics