Abstract

ALZATE-ZULUAGA, Norbey Yovany et al. Time-Frequency characteristics analysis for forecasting time series of particulate matter using Support Vector Regression and Particle Swarm Optimization. Rev.EIA.Esc.Ing.Antioq [online]. 2020, vol.17, n.34, pp.161-175.  Epub Sep 03, 2021. ISSN 1794-1237.  https://doi.org/10.24050/reia.v17i34.1347.

Atmospheric pollution by particulate matter is a problem recognized worldwide as a major risk factor for human health, over last years different models based on artificial intelligence has been proposed to forecast particulate matter concentration with the purpose of generate early warning systems that avoid people exposition. This paper analyzed a characterization scheme in time-frequency domain using the Wavelet to predict time series of PM10 and PM25 using the Support Vector Regression optimized with Particle Swarm Optimization (SVR-PSO). This paper also evaluated the effect of data imputation over estimations. Results showed that using time characteristics along with time-frequency characteristics SVR-PSO reach its best performance, also, it was found that use of data imputation does not affect SVR-PSO performance. The system proposed in this paper allow to estimate PM10 and PM2 5 concentrations with less error through time-frequency characteristics, in addition, it is capable to operate robustly against missing data, which improve its viability to be implemented in real scenarios.

Keywords : SVR; PSO; Wavelet Transform; Data imputation; Prediction; Regression.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )