20190726_Artificial neural network models for prediction of PM10 hourly concentrations, in the Greater Area of Athens, Greece




1. Title, Journal and Authors


Title : Artificial neural network models for prediction of PM10 hourly concentrations, in the Greater Area of Athens, Greece


Journal : Atmospheric Environment 40 (2006) 12161229


Authors : G. Grivas, A. Chaloulakou*


* School of Chemical Engineering, National Technical University of Athens, Heroon Polytechniou 9, 157 80 Zografos, Athens, Greece


2. Summary


 This study was conducted in order to investigate the potential of artificial neural network(ANN) methods, as tools for the prediction of PM10 hourly concentration, in the Greater Area of Athens, Greece. 4 sites with different characteristics were selected. The study was based on data measured for the two-year period. The meteorological variables were temperature, relative humidity, wind speed, wind direction, barometric pressure, solar radiation and amount of rainfall.


The concise data set was randomly divided in separate subsets. Three fourths data was conducted by training of neural networks, while the remaining cases where equally divided to validation and test sets. The early stopping technique was conducted in other to avoid over-fitting.


Three NN models were used. The first used the full set of the input variables, the second used the variables selected by a genetic algorithm optimization procedure and the third was developed without meteorological input variables. Networks with one and two hidden layers were evaluated.


In this study, genetic algorithms were used for dealing with noisy data, which are represented by binary strings of input variables. Also, Multiple linear regression models were considered as a reference for comparison with the NN models.


As a result, it was believed that ANNs should be considered as a serious candidate method for operational use in forecasting of hourly PM10 concentrations. The performance of examined NN models was superior in comparison with multiple linear regression models that were developed in parallel.



3. Contact


Jeongwoo Moon (Intern student)


Environmental Systems Engineering Lab.

School of Environmental Science & Engineering

Gwangju Institute of Science and Technology

1 Oryong-dong Buk-gu Gwangju, 500-712, Korea


Phone : +82-10-9384-8271

Office : +82-62-715-2475

E-mail : jeongwoomoon@gist.ac.kr





첨부 (0)