ASFMixer: Adaptive Segmented Fusion model for multivariate time series forecasting

Multivariate time series forecasting faces challenges like non-stationarity and complex dependencies. Existing models often ignore inherent periodic patterns and fail to adapt dynamically. We propose ASFMixer, an Adaptive Segmented Fusion Mixer that identifies key turning points to split series into variable-length sub-blocks. Each sub-block is processed by a dynamically selected model based on its trend, balancing local details and global patterns. Experiments show ASFMixer excels in long-term forecasting, offering a robust solution for handling fluctuating time series data.