This paper describes a proposed Dynamic Multi-Scale Coordination Framework (DMSC) to address the challenges of modeling complex time-series dependencies. DMSC addresses the challenges of static decomposition strategies, fragmented dependency modeling, and inflexible fusion mechanisms by incorporating the Multi-Scale Patch Decomposition block (EMPD), the Triad Interaction Block (TIB), and the Adaptive Scale Routing MoE block (ASR-MoE). EMPD dynamically partitions sequences into hierarchical patches through input adaptive patch coordination. TIB jointly models intra-patch, inter-patch, and inter-variable dependencies at each layer. ASR-MoE dynamically fuses multi-scale predictions using time-aware weights. Experiments on 13 real-world benchmarks demonstrate DMSC's leading performance and computational efficiency.
