Self-transformers: long-term series forecasting based on smooth series

This paper investigates the problem of long-term forecasting of time series. Previous Transformer-based models use various self-attention mechanisms to find remote dependencies. Autoformer designs a novel decomposition architecture with autocorrelation mechanisms. However, their models are less effective in dealing with long time series. In this paper, we propose a new time series forecasting model, Self-Transformers, based on Non-stationary Transformers. Designed to handle time series data with complex nonlinear trends and seasonality, our model improves forecasting performance and reduces the need for manual feature engineering. We validate the effectiveness of the model by conducting experiments on several benchmark datasets, and the results show that our model achieves significant performance gains on time series forecasting tasks.