PM2.5 Prediction Based on the Combined EMD-LSTM Model

Most of the research, conducted to date on the prediction of Fine Particulate Matter with a diameter less than 2.5 micrometers (PM2.5), based on machine learning and deep learning techniques, ignores the fact that the PM2.5 values are constantly changing over time. Although many researchers use Long Short-Term Memory (LSTM) neural networks based on time series to predict PM2.5 values, due to the instability of data, the results often had a certain lag. This paper
^♠
proposes to use the combined Empirical Mode Decomposition (EMD)—LSTM fusion model for the prediction of PM2.5 values. To evaluate the performance of the model in comparison to other existing models, experiments were conducted with a public PM2.5 data set, using the root mean square error (RMSE) and mean absolute error (MAE) as metrics. The results confirm the superiority of the combined EMD-LSTM model.