PCA/DFT como herramienta de pronóstico para series temporales de absorbancia registradas mediante captores UV-Vis en sistemas de saneamiento urbano

The purpose of this work is to introduce a forecasting method for UV-Vis spectrometry time series that combines principal component analysis (PCA), the discrete Fourier transform (DFT) and the inverse fast Fourier transform (IFFT). The corresponding absorbance time series were used for three different study sites: (i) Salitre wastewater treatment plant (WWTP) in Bogotá; (ii) Gibraltar pumping station in Bogotá; and (iii) San Fernando WWTP in Itagüí (in the southern part of Medellín). Each of these time series had an equal number of samples (5705). By reducing the absorbance time series dimensionality with PCA, 3, 5 and 6 principal components were used for each study site respectively; these altogether explain more than 97% of the variability. It was used the most important harmonic given by the DFT and the IFFT procedure, which removed from one to half values of the time series length. Therefore, forecast errors between 0,01% and 34% for 95% of the cases were obtained for the three study sites and the wavelength ranges (UV, Vis and UV-Vis). However, for 100% of the cases errors were lower than 37%, independently of the wavelength and the forecasting time.