Predicting PM10 and PM2.5 concentration in container ports: A deep learning approach

This study aims at predicting the concentrations of particulate matter in container ports. Meteorological data, terminal operation data, and data on PM2.5 and PM10 and other air pollutants at container ports were collected from Gwangyang Port in South Korea. A prediction model was developed using neural network methods such as recurrent neural networks (RNN), long short-term memory (LSTM), and multivariate linear regression (MLR). This study revealed that performance of LSTM was the highest. In addition, the performance of models with operating data is higher than the models without operating data as they have lower error values and stable decreasing patterns in a loss curve for training and validation loss. The proposed model could be used to provide PM information in advance to port workers and the public living in port cities so they can respond with personal hygiene and workplace health protection measures according to the predicted amount of PM.