An ensemble NLSTM-based model for PM2.5 concentrations prediction considering feature extraction and data decomposition

Fine particulate matter (PM2.5) is a hazardous air pollutant with an aerodynamic diameter of 2.5 μm or less, which can lead to severe health impacts such as cardiovascular disease, respiratory illnesses, and various types of cancer. Therefore, accurate forecasting of PM2.5 concentrations is crucial for public health and policy-making. However, due to the stochastic nature of PM2.5, achieving high prediction accuracy and efficiency remains a challenge. To address this challenge, this study proposes a hybrid deep learning model consisting of principal component analysis (PCA), discrete stationary wavelet transform (DSWT), and Nested LSTM (NLSTM) neural network to predict PM2.5 concentrations. The proposed model aims to leverage the strengths of each technique to achieve better accuracy and efficiency in PM2.5 forecasting. Specifically, PCA is employed as the feature extraction method to reduce the dimensionality of the data and improve computing efficiency. Additionally, DSWT is utilized to decompose the reduced-dimensional data into several sub-signals that are more regular and stable, enabling the NLSTM network to learn each sub-signal separately. Finally, the predicted values of each sub-signal are reconstructed to obtain the final PM2.5 forecast. The proposed model is validated using daily air pollutants and meteorological variables collected in Taiyuan, China, from January 1, 2016, to December 31, 2020. The long-term, medium-term, and short-term forecast results demonstrate that the proposed model achieves better accuracy and efficiency compared to existing models. Overall, the proposed hybrid deep learning model provides a promising solution for accurate and efficient forecasting of PM2.5 concentrations, and the findings of this study have important implications for public health and environmental policy.