Stabilized Long Short Term Memory (SLSTM) model: a new variant of the LSTM model for predicting ozone concentration data

Abstract

Ozone concentration prediction is essential for managing air quality. Thus, our study develops the Stablized-long short-term memory (S-LSTM) model to predict one-day-ahead ozone concentration in the Khorasan province, Iran. The S-LSTM model has processing units that efectively control fow information. In addition, the S-LSTM model normalizes the outputs of its memory, allowing it to process large data sets. Our study uses meteorological parameters and pollutant variables to predict one-day-ahead ozone concentration. Results show that the S-LSTM model outperforms the standard LSTM model in forecasting ozone concentrations. The S-LSTM model improved the values of Kling–Gupta Efciency (KGE), uncertainty at 95%, and Legates and McCabe Index (LMI) of the LSTM model by 9.4%, 73%, and 12%, respectively. The S-LTM model also improves the convergence speed of the LSTM model, which is important for modeling. Thus, the S-LSTM model can be utilized to monitor the concentrations of diferent atmospheric pollutants.