Accurate Multi-Site Daily-Ahead Multi-Step PM2.5 Concentrations Forecasting Using Space-Shared CNN-LSTM

Accurate multi-step PM2.5 (particulate matter with diameters ?2.5um) concentration prediction is critical for humankinds’ health and air population management because it could provide strong evidence decision-making. However, very challenging due to its randomness variability. This paper proposed a novel method based on convolutional neural network (CNN) long-short-term memory (LSTM) space-shared mechanism, named CNN-LSTM (SCNN-LSTM) multi-site daily-ahead forecasting self-historical series. The SCNN-LSTM contains multi-channel inputs, each channel corresponding one-site historical In which, CNN LSTM are used extract site's rich hidden feature representations in stack mode. Especially, the short-time gap patterns; mine features long-time dependency. Each extracted merged as comprehensive future forecasting. Besides, mechanism implemented by multi-loss functions achieve space information sharing. Therefore, final fusion of gap, dependency, information, which enables more accurately. To validate method's effectiveness, authors designed, trained, compared various leading methods terms RMSE, MAE, MAPE, R2 four real-word data sets Seoul, South Korea. massive experiments proved that accurately forecast only using time series running once. Specifically, obtained averaged RMSE 8.05, MAE 5.04, MAPE 23.96%, 0.7 four-site daily ahead 10-hour