Prediction of a High Concentration of PM2.5 near a Tailings Pond

In this paper, an error-correcting ensemble model combining empirical mode decomposition and reconstruction combined with statistical methods is proposed to improve the prediction accuracy of a high concentration of particulate matter with diameters of 2.5 mu m or less (PM2.5) around a tailings reservoir. The proposed prediction model consists of a prediction model of a high concentration of PM2.5 and an error correction model. The optimal prediction model is obtained by comparing the two indicators of fitting effect and result error, constantly adjusting the batch size and the number of network layers. The measured data of the tailings reservoir in Ma'anshan City, China, were used for validation. The results showed that the mean absolute error (MAE) of our long short-term memory-improved empirical mode decomposition-long short-term memory (LSTM-IEMD-LSTM) mode was 0.125, 0.661, and 4.372 lower than that of LSTM, multivariable linear regression-improve empirical mode decomposition-multivariable linear regression (MR-IEMD-MR), and autoregressive integrated moving average model-improve empirical mode decomposition-autoregressive integrated moving average model (ARIMA-IEMD-ARIMA), and the root-mean-square error (RMSE) of our LSTM-IEMD-LSTM compared with LSTM, MR-IEMD-MR, and ARIMA-IEMD-ARIMA decreased by 13.2%, 45.5%, and 84.2%. The model is also used to predict other high-concentration pollutants to test their generalization ability, which also has high prediction accuracy.