Publication: Forecasting of Carbon Monoxide Concentration Based on Sequence-to-Sequence Deep Learning Approach
No Thumbnail Available
Date
2021
Authors
Zaini N.
Ean L.W.
Ahmed A.N.
Journal Title
Journal ISSN
Volume Title
Publisher
Springer Science and Business Media Deutschland GmbH
Abstract
Carbon monoxide (CO) is one of the dangerous air pollutants due to its negative impact on human health. Therefore, accurate forecasting of CO concentration is essential to control air pollution. This study aims to forecast the concentration of CO using sequences to sequence models namely convolutional neural network and long short-term memory (CNN-LSTM) and sequence to sequence LSTM (seq2seq LSTM). The proposed forecasting models are validated using hourly air quality datasets from six monitoring stations in Selangor to forecast CO concentration at 1�h to 6�h ahead of the time horizon. The performances of proposed models are evaluated in terms of statistical equations namely root mean square error (RMSE), mean square error (MAE) and mean percentage error (MAPE). CNN-LSTM and seq2seq LSTM model excellently forecast air pollutant concentration for 6�h ahead with RMSE of 0.2899 and 0.2215, respectively. Additionally, it is found that seq2seq LSTM has slightly improved CNN-LSTM indicates the effectiveness of the architecture in the forecasting. However, both proposed architectures illustrate promising results and are reliable in the forecasting of CO concentration. � 2021, Springer Nature Switzerland AG.
Description
Air quality; Brain; Carbon monoxide; Errors; Forecasting; Mean square error; Network architecture; Air pollutants; Carbon monoxide concentration; Convolutional neural network; Deep learning; Forecasting models; Human health; Learning approach; Monitoring stations; Root mean square errors; Sequence models; Long short-term memory