Updated Prediction of Air Quality Based on Kalman-Attention-LSTM Network

Abstract

The WRF-CMAQ (Weather research and forecast-community multiscale air quality) simulation system is commonly used as the first prediction model of air pollutant concentration, but its prediction accuracy is not ideal. Considering the complexity of air quality prediction and the high-performance advantages of deep learning methods, this paper proposes a second prediction method of air pollutant concentration based on the Kalman-attention-LSTM (Kalman filter, attention and long short-term memory) model. Firstly, an exploratory analysis is made between the actual environmental measurement data from the monitoring site and the first forecast data from the WRF-CMAQ model. An air quality index (AQI) was used as a measure of air pollution degree. Then, the Kalman filter (KF) is used to fuse the actual environmental measurement data from the monitoring site and the first forecast results from the WRF-CMAQ model. Finally, the long short-term memory (LSTM) model with the attention mechanism is used as a single factor prediction model for an AQI prediction. In the prediction of O3 which is the main pollutant affecting the AQI, the results show that the second prediction based on the Kalman-attention-LSTM model features a better fitting effect, compared with the six models. In the first prediction (from the WRF-CMAQ model), for the RNN, GRU, LSTM, attention-LSTM and Kalman-LSTM, SE improved by 83.26%, 51.64%, 43.58%, 45%, 26% and 29%, respectively, RMSE improved by 83.16%, 51.52%, 43.21%, 44.59%, 26.07% and 28.32%, respectively, MAE improved by 80.49%, 56.96%, 46.75%, 49.97%, 26.04% and 27.36%, respectively, and R-Square improved by 85.3%, 16.4%, 10.3%, 11.5%, 2.7% and 3.3%, respectively. However, the prediction results for the Kalman-attention-LSTM model proposed in this paper for other five different pollutants (SO2, NO2, PM10, PM2.5 and CO) all have smaller SE, RMSE and MAE, and better R-square. The accuracy improvement is significant and has good application prospects.