MMST-LSTM: Leveraging Radar Echo Prediction for Emerging Consumer Applications in Edge Computing

With the increasing frequency of extreme weather events, there is a growing demand from the public for rapid and accurate short-term heavy precipitation forecasts. This study proposes a lightweight deep learning model, MMST-LSTM, which integrates Multiscale Context Feature Fusion Mechanism (MCFFM) and Mixed-Domain Attention Fusion Mechanism (MAFUM). While maintaining high prediction accuracy, MMST-LSTM significantly improves forecast speed. The MMST-LSTM model is particularly suitable for deployment in Mobile Edge Computing (MEC) environments, enabling fast localized forecasting. Experimental results demonstrate MMST-LSTM’s excellent predictive performance on two radar echo datasets, particularly in rapid response and handling localized data. Moreover, leveraging Smart Data-Driven Modeling (SDDM) technology with consumer-generated data enhances its application potential in smart consumer electronics products, providing an efficient tool for disaster weather alerts. This study introduces an innovative meteorological forecasting method and provides robust technical support for accurate weather warning systems, offering consumers timely and reliable weather information. This enables them to make more informed decisions, effectively reducing the potential risks and economic losses caused by extreme climate events.