An Improved Transformer Few Shot Fault Diagnosis Method for Integrated Navigation Based on Generative Adversarial Network

In practical application, it is difficult to effectively collect fault information of integrated navigation system. Aiming at the problem of few fault samples in INS/GNSS navigation system, an unsupervised integrated navigation system fault diagnosis method (shot for GA-TRANFD) is proposed in this paper. The method uses unsupervised training style, which only needs the normal integrated navigation data to train the fault diagnosis model, so that effectively solves the few-shot data problem. GA-TRANFD enhances the robustness of feature extraction by using a pyramid structure feature extractor, and realizes sequence reconstruction and fault extraction by applying Transformer Network with the idea of generating antagonism, and realizes the function of fault diagnosis. Finally, a published UCR time series data set and integrated navigation hardware-in-the-loop simulation data are used to verify the model. Taking the F1 and ROC scores as the evaluation metrics, which are commonly used in the classification field to represent the accuracy of classification, the experiment shows that the improved model improves the F1 and ROC scores of the references by 20.7%.