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

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