Research on Air Quality Prediction Model Based on CEEMDAN-CNN-BiLSTM-Attention

Air pollution is an increasingly serious problem with extensive impacts on human health and the environment. In order to improve the prediction accuracy, this paper proposes a hybrid model based on CEEMDAN-CNN-BiLSTM-attention mechanism for multi-step prediction of air quality time series. Firstly, complete ensemble empirical modal decomposition (CEEMDAN) is used to decompose the raw air quality data into multiple intrinsic modal functions (IMFs) in order to extract the time series features with different frequencies and effectively remove the noise and nonlinear components from the data. Then, a convolutional neural network (CNN) is used for feature extraction of the decomposed IMFs to capture the spatial features of the air pollutant data. Third, a bidirectional long and short-term memory network (BiLSTM) further extracts the temporal features of the time series and predicts the future pollutant concentrations in combination with the historical data. This paper introduces an attention mechanism into the BiLSTM model, which automatically focuses on the most important time periods for prediction by assigning different weights to the features at different moments, thus improving the sensitivity and prediction accuracy. Experimental results show that our model outperforms traditional models, such as RNN, LSTM and CEEMDAN-BiLSTM models, in air quality prediction.