PM2.5 Pollution Prediction Service based on Mogrifier LSTM Model

Accurate PM2.5 concentration forecasting service is essential for reducing the impacts of this air pollution. In this paper, we propose a novel PM2.5 pollution prediction service to achieve accurate PM2.5 forecasting effectiveness, which well integrates MLSTM (Mogrifier Long Short-Term Memory) prediction model and feature attention mechanism. Specifically, the MLSTM model captures long-term dependencies in PM2.5-related serial data through richer contextual interactions, while the feature selection module uses the attention mechanism to increase the overall prediction accuracy by allocating weights to each feature. We design representative experiments and the results show that compared to quintessential baseline approaches, our PM2.5 pollution prediction service achieves superior prediction accuracy.