Unlocking the Potential of Convolutional Dynamics: A Novel Approach to Time Series Forecasting

Time series forecasting is a critical component of decision-making across diverse domains, where precision in predictions is paramount for informed strategies. In this study, we introduce UPCD (Unlocking the Potential of Convolution Dynamics), a novel time series forecasting model that leverages convolutional structures and innovative techniques. UPCD excels in multi-variable long-term prediction tasks, outperforming both linear-based and transformer-based models in terms of accuracy. Our contributions include the introduction of NLMA Decomp (Non-Local Mean Attention for Seasonal-Trend Decomposition), a novel time series decomposition method that takes into account both local and global information for trend extraction, as well as the integration of a residual convolution module for effective handling of seasonal components. We conducted extensive experiments on nine real-world datasets, demonstrating UPCD’s state-of-the-art performance compared to other models.