Applying Time Series Analysis for Artificial Intelligence Based Particulate Matter Prediction

Artificial intelligence based prediction models provide good air pollution forecasters, proper to real time forecasting systems. Among them, artificial neural networks are the most used ones, being universal approximators. Usually, the identification of the best neural model (i.e. most accurate one) is based on experiments and results of time series analysis. The paper focuses on time series analysis for particulate matter (PM) air pollutant prediction with artificial neural networks in the Ploiesti city. Two types of neural models were used: feed forward and radial basis function. For each model we have experimented several architectures in order to identify the most accurate one in terms of root mean square error and average square error. The experimental datasets include five time series with concentration measurements of five air pollutants, PM10, CO, NO2, NOx, and SO2 in the period 2008-2012 at PH-6 air quality monitoring station from the Ploiesti city.