MVG-KAN: Multi-View Geo-Wind Guided KAN for PM2.5 Forecasting

Researchers propose MVG-KAN, a model for accurate short-term PM2.5 forecasting that addresses the limitations of existing methods in capturing complex pollutant dispersion driven by meteorological factors.

• The model separates stable daily and weekly patterns from non-periodic residual variations using a periodic-residual forecasting backbone.
• A Geo-Wind Graph combines geographic distance decay with wind-direction and speed to create a directed spatial prior for pollutant transport.
• A temporal Kolmogorov-Arnold network (TKAN) residual head learns nonlinear autoregressive corrections from de-periodized residuals and historical multi-pollutant sequences.

This approach enhances the modeling of local residual inertia and pollutant co-variation, providing a comprehensive representation of heterogeneous factors for improved air quality prediction.