Three-phase series arc fault detection based on two-dimensional attention PoolFormer

Abstract

Series arc faults are a major causation of electrical fires. The complexity of load types in low-voltage distribution systems makes arc faults detection more challenging for three-phase circuits with inverters. To solve this problem, this paper proposes a detection method based on two-dimensional attention PoolFormer. Firstly, a low-voltage three-phase series arc faults data acquisition platform is built to collect the required data. The collected current signals are encoded as pictures through image mapping and projected into a more discriminative space, while increasing the magnitude of the dataset. Subsequently, the two-dimensional attention PoolFormer algorithm model is constructed to fully exploit the feature information between different fault categories. This model has multi-scale parallel convolution to extract features of input samples and perform information fusion. Considering also the ability to seize the location characteristics of fault information well, the two-dimensional attention is designed to be added inside the algorithm, to grasp the precise location information to enhance the performance of the algorithm. Finally, the dataset is fed into the two-dimensional attention PoolFormer model for training and testing. The results show that the accuracy of the method proposed in this paper can achieve 99.36%.