Developing a wavelet-AI hybrid model for short- and long-term predictions of the pollutant concentration of particulate matter10

Abstract

Prediction of the particle concentration has received a great deal of research interest in recent decades, especially in the cities exposed to dust storms; hence, the use of early warning systems based on more accurate predictions to inform about the air quality situation in the next hours is very important. Thus, this study was conducted to improve the prediction accuracy of the particulate matter of aerodynamic diameter ≤ 2.5-μm particle concentration in the short term and long term, by combining the discrete wavelet transform technique with the artificial intelligence methods, i.e., adaptive neuro-fuzzy inference system, support vector regression, and artificial neural network. The data of the concentration of the suspended particles in Yazd city regarding 16 influential parameters were used during 2015–2019 in 6 models with and without wavelet, and the results obtained from these models were assessed and compared for hourly (short-term) and daily (long-term) timescales. The results based on 6 evaluation and validation criteria showed that the wavelet-support vector regression and wavelet-adaptive neuro-fuzzy inference system models have higher accuracy than other models in daily forecasting with an accuracy of about 99% and wavelet-support vector regression has higher accuracy in hourly forecasting with an accuracy of about 96%. The models combined with the wavelet transform produce better results rather than the single models showing the high effect of wavelet transformation in increasing the accuracy of artificial intelligence models used for prediction of the particulate matter ≤ 10 μm concentration.