Traditional time series analysis methods often face problems such as missing data and noise interference, which affects the predictive ability and accuracy of the model. Generative Adversarial Networks (GANs) have made significant progress in the fields of image generation and data restoration due to their excellent generative capabilities. In this study, a feature extraction and reconstruction method based on generative adversarial network (GAN) for time series data is proposed. By improving the structure of generative adversarial network and introducing the synergistic loss function of global optimization objective and single-channel optimization objective, combined with the Transformer architecture, the CTTS-GAN model is proposed. The model can better retain the features of the original data and improve the diversity of the data when generating time series data. The experimental results show that the data generated by CTTS-GAN exhibits a lower error (0.42) under the maximum mean difference (MMD) metric, showing a distribution closer to the real data. In addition, CTTS-GAN obtained better classification results than the traditional GAN model when using the generated data for the classification task, with a TSTR score of 0.83 on the Support Vector Machine classifier, which is significantly higher than other methods. This indicates that CTTS-GAN has a strong potential for application in generating high quality time series data.
