Risk assessment of aviation DC series arc based on reconstructed CBAM-CNN

The hazards of sustained arc and un-sustained arc are different. However, during the stage of arc development, there is a lack of effective methods to identify them, which is not conducive to the timely accurate assessment of arc risk. Therefore, this paper proposes a risk assessment method for aviation DC series arc based on a reconstructed CBAM-CNN. First, in the process of generating the feature set, a feature evaluation function is defined to screen the features. Then the existing convolution block attention module (CBAM) is improved by adding a reshaped layer and redefining spatial attention, which results in the reconstructed CBAM-CNN. Finally, the reconstructed CBAM-CNN takes the feature set as its input and output arc risk assessment results on the basis of enhancing the attention of important features. The validity of the reconstructed CBAM-CNN method is verified on an aviation DC arc generation platform. It is found that the proposed method has a higher training efficiency and evaluation accuracy than the CNN method and CBAM-CNN method. In addition, the reconstructed CBAM-CNN involves fewer parameters to be measured, which can reduce its dependence on computing resources.