Forecasting indoor pollutants concentrations using Fast Fourier Transform (FFT) and Regime Switching Models
Résumé
This paper considers the possibility that the high-frequency formaldehyde concentration follows a regime switching process. It explores the calibration of models built on Threshold Autoregression (TAR) combined with Fast Fourier Transform (FFT) for short-term forecasting of indoor formaldehyde fluctuation. The methodology uses signal decomposition with filtering of the original raw time series into different FFT components removing some high frequencies. Results from FFT-TAR model could predict HCHO concentration patterns with acceptable accuracy and suggest that there is benefit by taking choice of "cutoff" frequency. The mean absolute percentage error (MAPE) of the best scheme for 10-h (600); one day (1440) and 40h ahead out of sample forecasts, were obtained as follows: 3.54%, 14.36% and 15.64%.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...