Neurocomputing techniques to dynamically forecast spatiotemporal air pollution data
- PDF / 1,127,726 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 23 Downloads / 191 Views
ORIGINAL PAPER
Neurocomputing techniques to dynamically forecast spatiotemporal air pollution data Antonios Papaleonidas • Lazaros Iliadis
Received: 11 January 2013 / Accepted: 7 April 2013 Springer-Verlag Berlin Heidelberg 2013
Abstract Real time monitoring, forecasting and modeling air pollutants’ concentrations in major urban centers is one of the top priorities of all local and national authorities globally. This paper studies and analyzes the parameters related to the problem, aiming in the design and development of an effective machine learning model and its corresponding system, capable of forecasting dangerous levels of ozone (O3) concentrations in the city center of Athens and more specifically in the ‘‘Athinas’’ air quality monitoring station. This is a multi parametric case, so an effort has been made to combine a vast number of data vectors from several operational nearby measurements’ stations. The final result was the design and construction of a group of artificial neural networks capable of estimating O3 concentrations in real time mode and also having the capacity of forecasting the same values for future time intervals of 1, 2, 3 and 6 h, respectively. Keywords Artificial neural networks Machine learning Multi parametric ANN Pollution of the atmosphere Ozone estimation and forecasting 1 Introduction 1.1 Literature review The literature includes some interesting previous research efforts related to the use of artificial neural network (ANN) A. Papaleonidas (&) L. Iliadis Department of Forestry & Management of the Environment & Natural Resources, Democritus University of Thrace, 193 Pandazidou st., 68200 Orestiada, Greece e-mail: [email protected] L. Iliadis e-mail: [email protected]
towards forecasting air quality in the urban center of Athens. The main characteristic of these approaches is the fact that they are all developing a static overall network model which seems to be the optimal one. Also they have a seasonal nature by proposing models for specific times of the year (e.g. the summer months) and not for continuous long periods of time (Wahab and Al-Alawi 2002; Paschalidou et al. 2007; Iliadis et al. 2007; Ozcan et al. 2007; Ozdemir et al. 2008; Inal 2010; Paoli 2011). Moreover, these efforts have three basic disadvantages that motivated our research team to conduct this study. The first drawback is that they fail to work efficiently when necessary input data vectors are missing for quite long temporal periods. Regardless the credibility of modern measurement stations, the increased resources required for their technical support and also the random occurrence of unpredictable problems do not allow their absolute functionality. This frequently results in the lack of crucial data required for the proper risk assessment. The following Table 1 clearly supports this argument, by showing the percentage of data loss in five air pollution monitoring stations located in the wider urban center of Athens for the years 2003 and 2004 (Ministry of Environment, Energy & Climate Change 20
Data Loading...
