A Method for Detecting Abnormal Datas of Transfomer-Users Based on Deep Learning

The abnormal data in the distribution network adversely affects the load forecasting, scheduling, and troubleshooting processes of the power system, making it crucial to accurately detect and identify such data. This paper brings in a method for detecting unusual data in the distribution network using the Transformer model. First, an input sequence is constructed to capture various features, including time dependencies. Next, the Transformer model is incorporated into a generative adversarial network (GAN) to create the generator and discriminator, enhancing the detection capabilities for abnormal data through sequence reconstruction. Finally, the test data is fed into the trained model, with the discriminator generating an abnormality score. The power of the putted forward approach is validated through a true-earth situation study of a substation in Jiangsu. This research offers valuable insights into detecting abnormal data in power systems using deep learning techniques.