Abstract: An improved GAN-based anomaly identification model is designed for supply chain data. An exploratory analysis of the supply chain data was carried out using joint distribution and multi-paired sample Friedman test. The anomaly detection was performed with the data characteristics. To capture the temporal correlation of the data, LSTM was used as the base model for the generator and discriminator, and Cycle Consistency loss was used in the generator to prevent encoder and decoder conflicts, and Wasserstein loss was used in the discriminator to overcome the pattern collapse problem, while a non-parametric dynamic thresholding method was introduced for optimization, and thus identifying anomalies. The model was evaluated using accuracy, recall and F1 values and studied in comparison with the baseline method. The results show that the improved model is closer to the actual situation of supply chain data, can enhance supply chain flexibility and has high anomaly identification performance.