Fault diagnosis of ship power grid based on attentional feature fusion and multi-scale 1D convolution

The Ship Integrated Power System (SIPS) is evolving into a sophisticated network with prediction and active control functions, so accurate localization and identification of faults are crucial for the stable operation of the SIPS. The complex topology of power supply lines in ship power grids presents challenges in accurately locating and identifying faults. This paper presents a fault diagnosis model for the ship power grid based on attention feature fusion and multi-scale 1D convolutional neural network (AFF-MS-1DCNN), which can identify the fault type and locate the fault location only by using the three-phase currents of the busbar at the power supply output. By using the multi-scale 1DCNN, the method can effectively extract fault features on different scales. Furthermore, an attention mechanism is utilized to adaptively learn the weights of different features to enhance fault diagnosis precision. A transfer learning strategy is also applied to address variations in fault resistance. The experimental results demonstrate that the fault diagnosis accuracy of the AFF-MS-1DCNN model exceeds 98% under different fault resistance conditions, and it exhibits robust diagnostic performance even in the presence of noise interference.