Long-term time series forecasting plays a pivotal role in key application scenarios such as climate prediction, electricity load forecasting, and traffic flow assessment. Conventional single-architecture models—including Transformers, Convolutional Neural Networks (CNNs), and Multi-Layer Perceptrons (MLPs)—often struggle to comprehensively capture intricate temporal patterns. To address this limitation, we propose DSPNet, a carefully designed hybrid framework that synergistically combines multi-scale feature modeling with patch-based sequence representation. Although the individual components have been studied previously, to the best of our knowledge, their integration into a coherent pipeline—featuring bidirectional multi-scale mixing and a dedicated focus on local-global feature hierarchy—is novel. This design aims to fundamentally overcome the inherent limitations of single-architecture models in comprehensively capturing complex temporal patterns. The implementation of DSPNet is publicly available at https://github.com/jk16171216/DSPNet .
Liang et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: