A novel deep neural network for air quality prediction

Abstract

Air quality prediction is considered a major issue for public health, and early warning and control can reduce the bad impact imposed by air pollution on the health of residents. This paper aims to predict air quality by a dual-path multichannel deep neural network model. The model employs convolution layers, gate recurrent unit, and attention mechanism to learn spatial features, temporal features, and other key features to make an effective prediction. The dual-path structure learns different dimensions of data, namely the temporal dimension and the feature dimension. This allows the structure to use temporal dependencies and feature associations to build high-level features. Also, time series decomposition is applied to obtain trend, seasonality, and residual components in the monitored time series data. The proposed model takes as input a collection of pollutant data, meteorological data, and decomposed components in a window of 168 consecutive hours. Various experiments’ results show that our proposed system is superior to other systems in the accuracy of predicting air quality indices and pollutant concentrations.