Seasonal–trend-decomposed transformer has empowered long-term time series forecasting via capturing global temporal dependencies (e.g., period-based dependencies) in disentangled temporal patterns. However, existing methods design various auto-correlation or attention mechanisms in the seasonal view while ignoring the fine-grained temporal patterns in the trend view in the series decomposition component, which causes an information utilization bottleneck. To this end, a Transformer-based seasonal–trend decomposition methodology with a multi-scale attention mechanism in the trend view and a multi-view attention mechanism in the seasonal view is proposed, called DESTformer. Specifically, rather than utilizing the moving average operation in obtaining trend data, a frequency domain transform is first applied to extract seasonal (high-frequency) and trend (low-frequency) components, explicitly capturing different temporal patterns in both seasonal and trend views. For the trend component, a multi-scale attention mechanism is designed to capture fine-grained sub-trends under different receptive fields. For the seasonal component, instead of the frequency-only attention mechanism, a multi-view frequency domain (i.e., frequency, amplitude, and phase) attention mechanism is designed to enhance the ability to capture the complex periodic changes. Extensive experiments are conducted on six benchmark datasets covering five practical applications: energy, transportation, economics, weather, and disease. Compared to the state-of-the-art FEDformer, our model shows reduced MSE and MAE by averages of 6.5% and 3.7%, respectively. Such experimental results verify the effectiveness of our method and point out a new way towards handling trends and seasonal patterns in long-term time series forecasting tasks.

Multivariate time series modeling has been essential in sensor-based data mining tasks. However, capturing complex dynamics caused by intra-variable (temporal) and inter-variable (spatial) relationships while simultaneously taking into account evolving data distributions is a non-trivial task, which faces accumulated computational overhead and multiple temporal patterns or distribution modes. Most existing methods focus on the former direction without adaptive task-specific learning ability. To this end, we developed a holistic spatial-temporal meta-learning probabilistic inference framework, entitled ST-MeLaPI, for the efficient and versatile learning of complex dynamics. Specifically, first, a multivariate relationship recognition module is utilized to learn task-specific inter-variable dependencies. Then, a multiview meta-learning and probabilistic inference strategy was designed to learn shared parameters while enabling the fast and flexible learning of task-specific parameters for different batches. At the core are spatial dependency-oriented and temporal pattern-oriented meta-learning approximate probabilistic inference modules, which can quickly adapt to changing environments via stochastic neurons at each timestamp. Finally, a gated aggregation scheme is leveraged to realize appropriate information selection for the generative style prediction. We benchmarked our approach against state-of-the-art methods with real-world data. The experimental results demonstrate the superiority of our approach over the baselines.