A Hybrid Model Integrating LSTM with GARCH Family Models for the Ozone Concentration Prediction

To capture the characteristics of nonlinearity, high noise and volatility of the O
3
concentrations time series, some hybrid long short-term memory (LSTM) prediction models based on the GARCH family models (GARCH-LSTM) are proposed in this paper. The hourly air pollutant data and meteorological data of 9 monitoring stations in Shanghai are applied for model training and testing, and the models are compared with the LSTM, extreme gradient boosting (XGBOOST), support vector regression (SVR) and k-nearest neighbor (KNN) models. Experiment results show that our GE-LSTM model, a hybrid model combining the LSTM model with the GARCH model and EGARCH model, has the best prediction skills in terms of root mean square error and mean absolute error for the three-hour prediction tasks. The GE-LSTM model demonstrates outstanding performance in the O
3
concentrations prediction.