Air quality forecasting with hybrid LSTM and extended stationary wavelet transform

Air quality measurements and forecasting is one of the most popular research topics in the field of sustainable intelligent environmental design, urban area development and pollution control, especially for Asia developing countries, such as China. Deep learning (DL) technologies for time series data forecasting, such as the recurrent neural network (RNN) and long short term memory (LSTM) neural network, have attracted extensive attentions in recent years and have been applied to AQI forecasting. However, two problems exist in the literature. First, the volatility of the AQI data causes difficulties for singular DL models to produce reliable forecasting results. Second, a long history of the air-quality data is required in the training stage, which is usually unavailable. A novel forecasting model that integrates the extended stationary wavelet transform (ESWT) and the nested long short-term memory (NLSTM) neural network for PM2.5 air quality forecasting is proposed in this study. The results show that the proposed method outperforms state-of-art forecasting methods and recently published works in terms of different error metrics, such as absolute error, R-2, MAE, RMSE, and MAPE.

Automated summary

Air quality measurement and forecasting is a popular research topic in the field of sustainable intelligent environmental design, urban area development, and pollution control, especially for developing countries in Asia like China. This study proposes a novel forecasting model that integrates the extended stationary wavelet transform and nested long short-term memory neural network for PM2.5 air quality forecasting. The results show that the proposed method outperforms state-of-the-art forecasting methods in terms of different error metrics.