Academia.eduAcademia.edu

Figure 2 – uploaded by Joaquín Ordieres

See full PDFdownloadDownload figure
Fig. 2. Urban air quality monitoring network on the US—Mexico border in El Paso (Texas) and Ciudad Juarez, Chihuahua (Mexico).

Figure 2 Urban air quality monitoring network on the US—Mexico border in El Paso (Texas) and Ciudad Juarez, Chihuahua (Mexico).

Related Figures (11)

Fig. 1. U.S.—Mexican border region, as defined under the La Paz Agreement.
Fig. 1. U.S.—Mexican border region, as defined under the La Paz Agreement.
Primary NAAQS: the levels of air quality that the EPA judges necessary, with an adequate margin of safety, to protect the public health. Secondary NAAQS: the level of air quality that the EPA judges necessary to protect the public welfare from any known or anticipated adverse effects. National Ambient Air Quality Standard (NAAQS) for the US Table |
Primary NAAQS: the levels of air quality that the EPA judges necessary, with an adequate margin of safety, to protect the public health. Secondary NAAQS: the level of air quality that the EPA judges necessary to protect the public welfare from any known or anticipated adverse effects. National Ambient Air Quality Standard (NAAQS) for the US Table |
Input variables used in the forecast models Table 2
Input variables used in the forecast models Table 2
Typically, a neural network is composed of a set of neurons laid out in layers. Commonly, those layers are classified as input layer, hidden layers and output layer. with complex problems, particularly in the fields of association, classification and prediction. Many re- searchers have shown that neural networks can solve almost any problem more efficiently than the traditional modeling and statistical methods (Hornik, et al., 1989; Masters, 1993). In this paper, we compare three neuronal network architectures in order to obtain the best possible efficiency in the outcomes of Multilayer Perceptron (MLP), Square Multilayer Perceptron (SMLP), and Radial Basis Function (RBF) analyses.
Typically, a neural network is composed of a set of neurons laid out in layers. Commonly, those layers are classified as input layer, hidden layers and output layer. with complex problems, particularly in the fields of association, classification and prediction. Many re- searchers have shown that neural networks can solve almost any problem more efficiently than the traditional modeling and statistical methods (Hornik, et al., 1989; Masters, 1993). In this paper, we compare three neuronal network architectures in order to obtain the best possible efficiency in the outcomes of Multilayer Perceptron (MLP), Square Multilayer Perceptron (SMLP), and Radial Basis Function (RBF) analyses.
Fig. 5. Artificial neural network structure. Some neural networks do not have hidden layers and are used as more linear statistical techniques. These networks (with input and output layers only) are useful in many linear or semi-linear applications, but in general, it is difficult to get accurate results in non- linear problems (McCullagh and Nelder, 1989). We understand that particulate matter is clearly a non-linear
Fig. 5. Artificial neural network structure. Some neural networks do not have hidden layers and are used as more linear statistical techniques. These networks (with input and output layers only) are useful in many linear or semi-linear applications, but in general, it is difficult to get accurate results in non- linear problems (McCullagh and Nelder, 1989). We understand that particulate matter is clearly a non-linear
Fig. 7. MAE index evolution as a function of the number of hidden nodes for the different variances assessed. Another is the Mean Absolute Error (MAE), de- fined as:
Fig. 7. MAE index evolution as a function of the number of hidden nodes for the different variances assessed. Another is the Mean Absolute Error (MAE), de- fined as:
Fig. 6. Radial Basis Function structure.
Fig. 6. Radial Basis Function structure.
Fig. 8. Daily times series of the measured and predicted concentrations of PM>.s.
Fig. 8. Daily times series of the measured and predicted concentrations of PM>.s.
Fig. 9. Scatter plots of the measured and predicted concentrations of PM) s.
Fig. 9. Scatter plots of the measured and predicted concentrations of PM) s.
Fig. 10. Histograms of the residuals for the different models assessed. A short time PM>.5 prediction model has been built taking into account a large number of samples from The MLP network provides acceptable predictions, in spite of the difficult environmental conditions of the location (i.e. even though PM, data were considered, samples still show hard peaks of inmissions seasonally, mainly due to the common presence of dus storms in this area). The SMLP network shows a very similar behavior, although a few more neurons in the hid necessary to obtain the same error found case. This fact has a direct influence on t den layer are in the former he sample size necessary for the right training, as previously mentioned. Finally, the RBF network shows the best the shortest training times and best st behavior, with ability. These results suggest that the widely used MLP should be replaced by the more convenient RBF net work.
Fig. 10. Histograms of the residuals for the different models assessed. A short time PM>.5 prediction model has been built taking into account a large number of samples from The MLP network provides acceptable predictions, in spite of the difficult environmental conditions of the location (i.e. even though PM, data were considered, samples still show hard peaks of inmissions seasonally, mainly due to the common presence of dus storms in this area). The SMLP network shows a very similar behavior, although a few more neurons in the hid necessary to obtain the same error found case. This fact has a direct influence on t den layer are in the former he sample size necessary for the right training, as previously mentioned. Finally, the RBF network shows the best the shortest training times and best st behavior, with ability. These results suggest that the widely used MLP should be replaced by the more convenient RBF net work.
Related topics:
EngineeringGeographyComputer ScienceAir QualityAirborne ParticulatesMultilayer PerceptronFine Particulate MatterNeural Network Modeling
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved