Does Long-Term Series Forecasting Need Complex Attention and Extra Long Inputs?

Does Long-Term Series Forecasting Need Complex Attention and Extra Long Inputs? Abstract--As Transformer-based models have achieved impressive performance on various time series tasks, Long-Term Series Forecasting (LTSF) tasks have also received extensive attention in recent years. However, due to the inherent computational complexity and long sequences demanding of Transformer-based methods, its application on LTSF tasks still has two major issues that need to be further investigated: 1) Whether the sparse attention mechanism designed by these methods actually reduce the running time on real devices; 2) Whether these models need extra long input sequences to guarantee their performance? The answers given in this paper are negative. Meanwhile, a gating mechanism is embedded into Periodformer to regulate the influence of the attention module on the prediction results. This enables Periodformer to have much more powerful and flexible sequence modeling capability with linear computational complexity, which guarantees higher prediction performance and shorter runtime on real devices. Furthermore, to take full advantage of GPUs for fast hyperparameter optimization (e.g., finding the suitable input length), a Multi-GPU Asynchronous parallel algorithm based on Bayesian Optimization (MABO) is presented. MABO allocates a process to each GPU via a queue mechanism, and then creates multiple trials at a time for asynchronous parallel search, which greatly reduces the search time. Experimental results show that Periodformer consistently achieves the best performance on six widely used benchmark datasets.