An Air Quality Prediction Model Based On CNN-BiLSTM-Attention

In recent years,with the rapid development of urbanization and industrialization in China,the air pollution problem has become more and more serious,which gradually affects people’s production and life.Because of the fluctuation and nonlinear characteristics of air quality,it is difficult to predict air quality through a simple mathematical model.Therefore,the establishment of an efficient and accurate model for air quality prediction has very important guiding significance for air pollution control and air quality improvement.In this paper,an improved CNN-BiLSTM-Attention air quality prediction model is proposed to predict the air quality in the next hour.The model consists of convolutional neural networks(CNN),improved long short-term memory(LSTM)and attention mechanism(Attention).CNN is used to extract the features of input time series data,and the important feature data is selected.BiLSTM(bi-directional Long short-term Memory)which consists of multiple forward and backward linked improved LSTM respectively is used to calculate the time series data to obtain the output data after feature extraction by CNN.LSTM is improved by introducing the 1-tanh function behind the forget gate of LSTM.After introducing the 1-tanh function,the output value of the forget gate can be in a more obvious range.Therefore,the improved LSTM can preserve more characteristics of the input data,and improve the learning ability of LSTM.Attention is used to capture the effect of characteristic conditions at different times on air quality prediction value,and the correlation between each data the time series and the predicted value is obtained.To prove the validity of the improved CNN-BiLSTM-Attention,the model,multi-layer perceptron(MLP),CNN,recurrent neural network(RNN),LSTM,BiLSTM,CNN-BiLSTM and CNN-BiLSTM-Attention improved CNN-BiLSTM-Attention are compared by experiments in this paper.All models use the same training set data and the optimal model of each model is saved after the training.All models are tested using the same test set data.By comparing the experimental results of eight models,the mean absolute error(MAE)and root mean square error(RMSE)of the improved CNN-BiLSTM-Attention model are both the smallest.The MAE is 6.205 and the RMSE is 9.616.R-square(R~2)is 0.9704,which is closest to 1.Therefore,the improved CNN-BiLSTM-Attention air quality prediction model is suitable for air quality prediction.It has high prediction accuracy for air quality and can provide an effective reference for relevant departments to take measures in advance to improve air quality.