Abstract: Since physical characteristics of arc fault are complex and there are loads with waveforms similar to arc fault in circuits, misjudgment of traditional arc fault detection methods is high. To improve the arc fault detection accuracy, a multi-sensor data fusion algorithm that includes an adaptive weighted fusion algorithm and a neural network fusion algorithm is proposed. Date is acquired with a temperature sensor, an acoustic sensor and a light sensor. The adaptive weighted fusion algorithm can overcome uncertain of single sensor, extract the arc fault characteristics of homogeneous sensors, and provide accurate test sample date for neural network fusion, making the probability of identifying arc fault more accurately by automatically adjusting the weights for all kinds of heterogeneous sensors. Experimental results show that the algorithm can effectively extract the arc fault characteristics and improve identification accuracy up to 98%, meeting the high accuracy arc fault detection requirement.

Key words: adaptive weight, arc fault, multi-sensor, neural network