Distribution Network Fault Identification Based on the Improved Transformer Network

Distribution network fault detection is crucial for ensuring the reliability and safety of power supply, minimizing downtime, and avoiding potential damages to the network and connected equipment. Generally speaking, there are two application scenarios which include waveform analysis and inspection image recognition. Due to the problems such as insufficient data, complex background and class imbalance, existing machine learning algorithms do not exhibit satisfactory performance. This paper proposes a unified framework based on Transformer network to deal with the above-mentioned problems. We mainly focus on better feature extraction in the two application scenarios. For example, we use statistical features as input in waveform analysis, while object relations are combined with the appearance features in inspection image recognition. Experiments based on field data prove that our method outperforms related works in terms of accuracy, efficiency and extensibility, which largely promotes the construction process of intelligent power grids.