ISBN: 978-981-18-7950-0 DOI: 10.18178/wcse.2023.06.051
Dual-channel Feature Fusion Model for Air Pollutants Forecast
Abstract—Air pollution is severely injurious to human health, and hence short-term prediction for various air pollutants concentrations is beneficial for early prevention and travel planning. To reduce the contradictory information between time trend and variable correlation when predicting, the dual-channel feature fusion model is proposed. The model first obtains the time-series characteristic through the temporal feature extraction module based on TPA-LSTM, then captures the variable relevance with the structural feature extraction module aggregated GNN and LSTM, and finally integrates the two features by a fusion gate to realize forecast. Comparative experiments on air pollutants dataset from the state-controlled air station in Jinan by different models reveal that our method demonstrates superior properties.
Index Terms—Air pollutants forecast, Multivariate time-series prediction, Feature fusion, LSTM, GNN
Yujie Li, Xu Qiao, Zhiping Liu, Rui Gao
School of Control Science and Engineering, Shandong University, CHINA
Cite: Yujie Li, Xu Qiao, Zhiping Liu, Rui Gao, "Dual-channel Feature Fusion Model for Air Pollutants Forecast" Proceedings of 2023 the 13th International Workshop on Computer Science and Engineering (WCSE 2023), pp. 343-348, June 16-18, 2023.