DTW-Enhanced CNN-GRU Model for PM2.5 Forecasting in Isfahan
A novel deep learning model that merges Dynamic Time Warping (DTW) with CNN-GRU architecture has been developed to provide dependable PM2.5 forecasts for over 48 hours in Isfahan, Iran, a city characterized by limited monitoring networks and intricate pollution behaviors. This approach employs DTW for historical sampling to detect comparable pollution trends among nearby stations, incorporates meteorological data, and is built for scalability, avoiding the need for resource-heavy transformer models or external simulation tools. This innovation fills a significant void in public health early-warning systems, addressing the challenges faced by current deep learning methods in maintaining prediction stability over longer timeframes.
Key facts
- The framework combines DTW for station similarity selection with a CNN-GRU architecture.
- It targets long-term PM2.5 forecasting beyond 48 hours.
- The study is set in Isfahan, Iran, a city with complex pollution dynamics and limited monitoring coverage.
- Three key innovations: DTW-based historical sampling, lightweight CNN-GRU with meteorological features, and scalable design.
- The method avoids computationally intensive transformer models or external simulation tools.
- Reliable long-term forecasting is critical for public health early-warning systems.
- Existing deep learning approaches struggle to maintain prediction stability beyond 48 hours.
- The framework is optimized for cities with sparse monitoring networks.
Entities
Locations
- Isfahan
- Iran