Constructing a PM2.5 concentration prediction model by combining auto-encoder with Bi-LSTM neural networks

Abstract Air pollution problems have a severe effect on the natural environment and public health. The application of machine learning to air pollutant data can result in a better understanding of environmental quality. Of these methods, the deep learning method has proven to be a very efficient and accurate method to forecast complex air quality data. This paper proposes a deep learning model based on an auto-encoder and bidirectional long short-term memory (Bi-LSTM) to forecast PM2.5 concentrations to reveal the correlation between PM2.5 and multiple climate variables. The model comprises several aspects, including data preprocessing, auto-encoder layer, and Bi-LSTM layer. The performance of the proposed model was verified based on a real-world air pollution dataset, and the results indicated this model can improve the prediction accuracy in an experimental scenario.