An air quality prediction model based on CNN-BiNLSTM-attention

In recent years, the air pollution problem has been aggravated, which has brought some problems to people's production and life. A simple mathematical model cannot accurately predict air quality because of the characteristic of air quality volatility and obvious nonlinear characteristics. Therefore, this paper proposes a CNN-BiNLSTM-Attention air quality prediction model to forecast the AQI in the next hour. In the model, convolutional neural networks (CNN) are used to extract characteristics from the input air quality data and meteorological data. NLSTM is an improvement on long short-term memory (LSTM) to make output value range of forget gate more accurate, thus preserving more characteristics of the data. BiNLSTM is used to predict the time series data. Attention mechanism (Attention) is used to capture the effect of characteristic conditions at imparity times on AQI prediction. To prove the accuracy of the model, CNN-BiNLSTM-Attention, recurrent neural network (RNN), LSTM, ARIMA, BiLSTM, CNN-BiLSTM, and CNN-BiLSTM-Attention are used to forecast the hourly AQI from 00:00 on January 1, 2020, to 23:00 on September 30, 2020, in Shijiazhuang, Hebei Province in this paper. The results show that the MAE of CNN-BiNLSTM-Attention is 5.987 and the RMSE is 9.231. They are the smallest. R2 is 0.9741, which is the closest to 1. The CNN-BiNLSTM-Attention air quality prediction model is more suitable to predict air quality, informs people in advance of air pollution who can take corresponding measures to reduce air pollution.