Topology recognition of substation grounding grids based on small-sample electromagnetic induction images

Abstract

To achieve intelligent recognition of the grounding grid topology structure, this paper addresses the challenge of limited measured samples, which makes it difficult for neural networks to be applied in grounding grid recognition. A classification system for grounding grid topology structure recognition is designed. By using the commercially available simulation software CDEGS, simulations are performed based on the parameters of the measured environment to quickly generate a large number of magnetic field intensity distribution images. Given the complex electromagnetic environment in substations and the presence of significant electromagnetic noise in the measured magnetic field intensity distribution images, a style transfer algorithm based on transfer learning is developed. This algorithm adds the electromagnetic noise from the measured images to the simulated images, thus creating a database of measured image samples. Based on the database of measured images, the AlexNet model, a deep learning algorithm for image recognition, is employed to investigate the grounding grid topology structure recognition method. This approach facilitates the extraction of topological structure features from the magnetic field intensity distribution images of the grounding grid, followed by classification and recognition. System test results demonstrate that the system achieves a high recognition rate for the measured images.