Presenter/Author Information

Kostas Karatzas
D. Voukantsis

Keywords

computational intelligence, periodograms, principal component analysis, self-organising map, artificial neural networks, air quality, atmospheric environment, quality of life

Start Date

1-7-2008 12:00 AM

Abstract

Air quality management is among the most challenging problems in terms of analysis and modelling. Air quality modelling and forecasting is directly affected by the highly nonlinear relationships between pollutants and weather, while in many cases there is insufficient domain knowledge due to the influences of local conditions. As atmospheric quality has an impact on the quality of life of millions of people, the ability to reveal interrelationships between parameters that influence environmental decision making is very important. In addition, forecasting of such parameters for the purpose of early warning and health risk prevention is of paramount importance for sensitive parts of the population. In the present paper a number of Computational Intelligence methods are presented with the purpose to the investigate atmospheric quality parameters, towards better understanding of the interrelationships between pollutants, and with the aim to improve forecasting of critical values. For this reason, the use of Fast Fourier Transformation for the construction of Periodograms is firstly presented, followed by the application of Principal Component Analysis. Then, Self Organizing Maps, a method based on the Neural Networks approach, is investigated and applied, for knowledge extraction and atmospheric parameter analysis. Last, an Artificial Neural Network based on the multi-linear perceptron (MLP) model is presented in order to construct prediction models. Results indicate a number of important features within the data investigated, and reveal hidden interrelations, thus providing valuable information for the understanding and the explanation of environmental problems, and for the support of environmental policy and decision making in both long and short terms. It is also demonstrated that the performance of forecasting models justify their selection for early warning information services.

Share

COinS
 
Jul 1st, 12:00 AM

Studying and predicting quality of life atmospheric parameters with the aid of computational intelligence methods

Air quality management is among the most challenging problems in terms of analysis and modelling. Air quality modelling and forecasting is directly affected by the highly nonlinear relationships between pollutants and weather, while in many cases there is insufficient domain knowledge due to the influences of local conditions. As atmospheric quality has an impact on the quality of life of millions of people, the ability to reveal interrelationships between parameters that influence environmental decision making is very important. In addition, forecasting of such parameters for the purpose of early warning and health risk prevention is of paramount importance for sensitive parts of the population. In the present paper a number of Computational Intelligence methods are presented with the purpose to the investigate atmospheric quality parameters, towards better understanding of the interrelationships between pollutants, and with the aim to improve forecasting of critical values. For this reason, the use of Fast Fourier Transformation for the construction of Periodograms is firstly presented, followed by the application of Principal Component Analysis. Then, Self Organizing Maps, a method based on the Neural Networks approach, is investigated and applied, for knowledge extraction and atmospheric parameter analysis. Last, an Artificial Neural Network based on the multi-linear perceptron (MLP) model is presented in order to construct prediction models. Results indicate a number of important features within the data investigated, and reveal hidden interrelations, thus providing valuable information for the understanding and the explanation of environmental problems, and for the support of environmental policy and decision making in both long and short terms. It is also demonstrated that the performance of forecasting models justify their selection for early warning information services.