DCT-based Least-Squares Predictive Model for Hourly AQI Fluctuation Forecasting

Abstract

Recently the issue of air quality has become a global public health concern. As air pollution has been reported as the largest single environmental health risk in the world, analysis and prediction of air quality is increasingly important. Normally, either statistical models or CTM (“deterministic chemistry-transportâ€) models are used for forecasting air PM (“particulate matterâ€) levels. Actually, hourly air quality fluctuation is also one time-series. Compared with commonly used deterministic photochemical air quality models, data-driven or time-series-based modeling is simpler and can also perform well even be more accurate. In this study, a called DCT(“discrete cosine transformâ€)-based least-squares predictive model is proposed for forecasting hourly AQI (“air quality indexâ€) from time-series analysis or data-driven modeling perspective. The proposed DCT-based predictive method is implemented in combination with the least-squares method to compute the called least-squares-optimum DCT coefficients for forecast modeling on the basis of finite hourly AQI observations. The proposed method yields one good result of average 93.24% predictive accuracy in forecast experiments at five monitoring stations in Xiangtan of China. Experimental results and analysis of performance comparisons of the proposed DCT-based least-squares predictive model with the classical BP-ANN model, the Fourier-series-based least-squares predictive model and the ARIMA model indicate that for the same tasks of forecasting hourly AQI fluctuations, the proposed DCT-based predictive model outperforms the former two competitive models and performs slightly better than or comparable to the ARIMA model. It is indicated that the hourly AQI fluctuations can be well forecasted by the proposed DCT-based least-squares predictive model with using about 12-term least-squares-optimum DCT coefficients.

Keywords: DCT-based model, forecasting, hourly AQI, least-squares fitting