Prediction of PM10 and TSP Air Pollution Parameters Using Artificial ...

Middle-East Journal of Scientific Research 14 (7): 999-1009, 2013 ISSN 1990-9233 © IDOSI Publications, 2013 DOI: 10.5829/idosi.mejsr.2013.14.7.2171

Prediction of PM10 and TSP Air Pollution Parameters Using Artificial Neural Network Autoregressive, External Input Models: A Case Study in Salt, Jordan 1

Mouhammd Alkasassbeh, 2Alaa F. Sheta, 3Hossam Faris and 4Hamza Turabieh Department of Computer Science, Mutah University, Jordan 2 Department of Computer Science, College of Computers and Information Technology, Taif University, Saudi Arabia 3 Department of Business Information Technology, The University of Jordan, Amman, Jordan 4 Department of Information Technology, College of Computers and Information Technology, TaifUniversity, Saudi Arabia 1

Abstract: Air pollution in large cities has been a major and a serious environmental problem all over the world; hence, many countries initiated air quality management systems to monitor and control pollution rates around big cities. It was found that harmful emission into the air is a symbol for environmental force that affects seriously man’s health, natural life and agriculture; thus leading to major loss on the nation’s economy. Government in industrialized countries deployed many regulations to apply restrictions on emission limits thus reduce the levels of pollution in air and enforce the international standards for air quality levels. The main objective of this study is to develop a non-parametric Artificial Neural Network (ANN) models to predict both the Particulate Matters (PM10) and Total Suspended Particles (TSP) in Salt, Jordan. A data set collected around Al- Fuhais cement plant over one-year period (2006-2007) by eight monitoring stations were used in our study. The proposed ANN models considered the meteorological parameters: Temperature (Temp), Relative Humidity (RH), Wind Speed (WS) as inputs. We developed two Artificial Neural Network based AutoRegressive with eXternal (ANNARX) Input models to provide high performance modeling for the PM10 and the TSP air pollution parameters. Experimental results show that ANNARX can provide good modeling results using a limited number of measurements. Key words: Air pollution

PM10

TSP

Artificial Neural Network.

INTRODUCTION

concentrations of PM10 for Bordeaux metropolitan area. The suggested nonlinear model was designed based on empirical relationships between the measured PM10 and other primary pollutants and meteorological parameters. Moreover, the Extended Kalman filter algorithm is used to estimate 1-day ahead prediction of the extended state, containing model parameters and daily mean PM10. This approach was applied using data from a monitoring site in Bordeaux (south France). Recently, Soft computing techniques such as ANN, Fuzzy Logic (FL) and Genetic Algorithms (GAs) were used in solving many air pollution model development problems. ANN successfully applied to atmospheric pollution problems. In [10], authors introduced a new Back-Propagation (BP) algorithm for the modeling of air

Air quality prediction helps significantly in the management of our environment. Predicting air quality is still a challenge because of the complexity of its processes and the strong coupling across many parameters, which affect the modeling process [1, 2]. In urban areas, human health has been affected by air pollution [3]. The most significant air pollutants are nitrogen dioxide, sulfur dioxide, polycyclic aromatic hydrocarbons (PAHs) and suspended particulate matter (PM) including dust, soot and smoke [4]. Many techniques were adopted to handle the air pollution forecasting problems [5–8]. In [9], an adaptive nonlinear state-space based modeling techniques have been designed to predict daily mean

Corresponding Author: M. Alkasassbeh, Computer Science Department, Mutah University, Jordan.

999

Middle-East J. Sci. Res., 14 (7): 999-1009, 2013

pollution time series. ANN was used to solve many forecasting problems for environmental systems [11, 12]. In [13] authors proposed a unified approach for integrating implicit and explicit knowledge in neuro-symbolic systems as a combination of neural and neuro-fuzzy modules [13]. There technique was used to extract fuzzy rules from neural implicit knowledge modules. Genetic Algorithms (GAs) was successfully used to extract the best features to develop an air-quality model in [14]. In this paper, we present our initial idea of developing ANN models for both the PM10 and the TSP air pollution parameters. The proposed ANN will consider a limited number of input variables, which include Temperature, Relative Humidity, Wind Speed to predict the values for both the PM10 and TSP. A BP learning algorithm is used to train the ANNARX models. Our study considered one of the most polluted areas in Jordan, which is Al-Fuhais City in Salt. It is characterized by the presence of the largest cement plant in the country. The product size amounts up to 2 million tons a year. In addition to a high volume of vehicular traffic transporting cement, the factory and the mine are situated in an area surrounded by residential districts. The particulate matter emitted from industrial processes has been of particular interest for the possible delayed health effects associated with continuous exposure of population. Meteorological Parameters: Many international environmental agencies like; Intergovernmental Panel on Climate Change (IPCC), United Nations Environment Programme (UNEP), Earth System Governance Project and Global Environment Facility (GEF) are responsible in protecting human health and the environment, so they have set up standards of air quality guidelines and some threshold for levels of these pollutants in the air, thus, when these levels exceed the threshold, it means that human health is in dangerous [15, 16]. Meteorology, along with emissions and atmospheric chemistry, is well known as a major contributor to air pollution episodes. The important meteorological parameters which influence air pollution can be classified into primary parameters (i.e. wind direction and speed, temperature, atmospheric stability and mixing height) and secondary parameters (i.e. precipitation, humidity, solar radiation and visibility). The parameter varies widely as a function of latitude, season and topography. Just as weather affects the severity of air pollution, air pollution, may in turn affect weather conditions [17].

Particulate Matters (PM10): Particulate matter, also known as particle pollution or PM, is defined as,”a complex mixture of extremely small particles and liquid droplets”. Particle pollution consists of a number of elements. They include acids (such as nitrates and sulfates), organic chemicals, metals and soil or dust particles. The size of the PM particles is normally small such that it can enter through either the throat or the nose to the human’s lung. These particles can affect the heart, lungs and cause serious health effects [3, 16]. Studies of long- term exposure to air pollution suggest an increased risk of developing various types of cancer [15, 18]. Particulate matter is measured and expressed as the mass of particles contained in a cubic meter of air (micrograms per cubic meter, or µg/m3). These particles are in 10 micrometres in diameter or smaller. The United States has established two types of PM air quality standards: PM10 and PM2.5. It was suggested that PM10 is a suitable parameter for monitoring PM pollution around industrial activities involving the manipulation of mineral dust (e.g. ceramic, cement or brick production, mostly linked to fugitive emissions) because the mass of dust is mainly in the size bigger than 2.5 µm fraction [19-21]. Total Suspended Particles (TSP): Total suspended particles are those particles which range in diameter from 0.1 to 10 microns and remain long suspended in the air [16]. The deposition rate is relatively slow and depends on the natural conditions of moisture, wind, heat and others. Suspended particles which could be dangerous air pollutant particles as possible to reach the lungs and settle there. These particles constantly enter the atmosphere from many sources such as: Manmade like; fuel combustion in car engines and power plants, factories, wood burning, industrial processes, power generation [22] or Natural side sources like; bacteria and viruses, salt particles from evaporating sea water, Pollen, Fungi, volcanic eruptions and many others [23]. These materials suspended in the air cause multiple health problems. Moreover, TSP causes other several problems, including increased emergency cases in the hospitals for heart diseases, lungs and sometimes early death. Air Pollution in Jordan: Jordan is divided into 12 governorates. They are Ajlun, Amman, Aqaba, Balqa, Irbid, Jerash, Kerak, Ma’an, Madaba, Mafraq, Tafilah and Zarqa as shown in Figure 1. In Table I, we provide

1000

Middle-East J. Sci. Res., 14 (7): 999-1009, 2013 Table 1: Statistics of Jordan Governorates According to 2008 Estimate [20] Province Total Population (2008 est.) Area(km2) Population Density (person/(km2) Ajlun 118,496 412 287.1 Amman 1,939,405 8231 246.3 Balqa 349,580 1076 324.9 Irbid 950,700 1621 570.3 Zarqa 838,250 4080 205.5

Urban Population (%) 67.4 91.4 63.9 76.4 95.3

Rural Population (%) 32.6 8.6 36.1 23.6 4.7

measured the submicron particle number concentrations in the urban/suburban atmosphere of Amman-Jordan during the spring of 2009. Their main objective was to distinguish the differences in the submicron particle number concentrations with/without dust episodes. Why Artificial Neural Networks?: ANN is a computational technique inspired by in the nervous systems of the human brain [28, 29]. ANN is a type of non-linear processing system that is perfectly appropriate for a widespread domain of applications [30, 31]. They are generally trained using some learning algorithm to memorize a certain function, classify patterns or take appropriate decision using a set of collected measurements from the application environment. There are two main features make neural networks a very useful tool in solving prediction and modelingproblems.

Fig. 1: Map of the Governorates of Jordan statistics about various governorates, including their areas, rural population, urban population and population density. Beginning of the 70’s, Jordan started to realize the pollution problems associated with air. The impact of air pollution began to deteriorate and affects the quality of air and as a result it has created health problems, which promoted the government to take action in controlling air emissions [24]. In 1985, an air pollution monitoring network for Amman was developed as part of a collaborative research program between the Royal Scientific Society of Jordan and Environment Canada, Pollution Measurement Division, Ottawa, Canada to help achieving national air quality standards and national emission standards for Jordan [25]. An early study of air pollution monitoring in Amman, Jordan was presented in [26]. Author provided some statistics about the TSP matter concentrations at four sampling sites form July 1986 to June 1987. Carbon monoxide, nitrogen dioxide, sulphur dioxide PM10 and total TSP were measured at various sites. In [24], authors provided a study of Black Carbon Levels in City Centers and Industrial Centers in Jordan. They measured the light absorption coefficients of black carbon at several urban and industrial locations in Jordan during the summer of 2007 and the winter of 2008 using the photo-acoustic instrument at a wavelength of 870nm. In [27], authors

Those features are; their ability to learn and generalize. Learning process depends on providing a set of training data, which used to adjust the network weights, using a described learning algorithm. After training process have been successfully achieved, the network will be able to recognize a certain output when a newly data is presented to its input layer. ANN found to be clever on generalization. They have the capacities to identify likenesses amongst diverse input patterns. ANN has been used both to estimate parameters of a formal model and learn to simulate the process model itself to predict future outcomes. They can deal with non-linear, nonstationary, or chaotic function. Neural networks have been successfully used to solve a variety of prediction and forecasting problems. It has proven to be efficient in number of applications [32-36]. ANN Architecture: ANN was introduced by Mcculloch and Pitts in 1943. They are parallel working systems that consist of a number of processing units interconnected in a network, called neurons. These neurons are grouped together to form a layer. Each neuron has a number of inputs and a single output, in our case study. Each input

1001


has an assigned factor or parameter called the weight. The way how a neuron is working is as follows: input signal to each neuron is multiplicity by the corresponding weight then the result from the multiplication is summed and passes through a transfer function, most likely to be a sigmoid function. If the result of the summation is over a certain threshold, the neuron output will be activated else the output is not. The Multilayer Perceptron: Processing units or neurons can be interconnected into a network in different fashions, the most popular one of these fashions is the Multi-Layer Perceptron (MLP) network and due to the structure of the MLP it is also named as multilayer Feed Forward (FF) network. This structure orders the units in layers, letting each unit in a layer to receive inputs only from preceding layer outputs or external inputs. If the MLP has two such layers of units, it is referred to as a two-layer network, if it has three layers; it is called a three- layer network and so on. Figure 2 depicts a one-layer feed forward MLP. The presented network is a fully connected because all inputs/units in one layer are connected to all units in the following layer; input, hidden and output layers. The depicted network consists of N inputs, one hidden layer and one output. The hidden layer hhas fully connected neurons with nonlinear sigmoid function and the output layer has neurons with linear sigmoid functions. To clarify Figure 2 mathematically, given the inputs xi of the layer n, the outputs yi of the layer n + 1 are computed as: = ui

+1 ∑ ( win, +j 1x j ) + win,bais

yi = f(ui)

(1) (2)

MLP may use different activation functions (AF). For example; Identity function f(u) = u

(3)

Logistic Function (Sigmoid) f (u ) =

(1 − e − (1 + e

− x

x

)

)

(4)

in fact this function is the default function for the MLP and mostly used.

Fig. 2: A fully connected one layer feed forward network Gaussian function f (u) = e

u2

(4)

MATERIALS AND METHODS Area of Study: The studying area was in Al-Fuhais city which has Al- Fuhais cement plant (21.5 km west of Amman). The area of the city is about 30 square kilo meters and total population about 20,000 Persons. The city located on hills varying in elevations ranging from between 550m to 1200m above mean sea level. The annual average temperature in the city ranges from 18-220c. The coldest month in the year is January, where the average temperature reaches 60c, while the warmest month in the year is August with the average temperature increases to reach 240c.The precipitation may reach 600mm. The average wind speed blowing throughout the year is 5.46 knot. Al-Fuhais containing cement plant, which considered the biggest cement factory in Jordan, as the product size is amounted to 2 million tons a year. The factory has two production lines; the first line stack is 67.5m in height and 3.64m in diameter, the second line stack is 86m in height and 3.1m in diameter. Dust is emitted from different cement manufacturing processes and activities, including mining, milling, transport, handling of raw materials and kiln operation. As well as, dust is also emitted from vehicles and trucks. Because Al-Fuhais cement plant is surrounded by residential areas causing adverse health impacts [37, 38]. Monitoring Instruments: Measuring process is carried out by using High Volume Sampler (HVS)GS3210and Beta attenuation monitors in order to measure the TSP and PM10 concentration. The used HVS is shown in Figure 3. In our case, the HVS is installed on the terrace of the buildings and the instrument was placed in a parallel position to the ground. The HVS was used as a vacuum cleaner; it drew a volume of air through a filter that captured suspended particles. The sampler is operated at

1002


flow rates of 1m3/min collected on pre-weighed glass fiber filters. The filters weighing as accurately as possible since the ambient humidity is the common cause of producing erroneous filter weights. It was run continuously for a period of 24 hours and at the end of the sampling period, the filter paper was removed and a daily reading for the instrument was taken. Distribution of Monitoring Stations: A monitoring program for PM10 and TSP was deployed in order to assess the environment situation in the area. The Faculty of Agricultural Technology, Al-Balqa Applied University (BAU), Salt, Jordan used eight monitoring stations to measure PM10 and TSP concentrations. The eight stationswere selected up to a distance of 7 km around the plant and cover different types of environments and surround the main industrial activities in the area. Five stations were used to measure the PM10 concentrations and the other three were used to measure the TSP concentrations. The air quality monitoring stations measuring PM10 named Deir El-Latin, Eshtaiwi Home, George Hassan Home, Bab Al- Salam Palace and Al-Kamaliyya School. The air quality monitoring stations measuring TSP named Al-Hashimiyya Palace, Bab Al-Salam Palace and Swaileh Municipality. In Figure 4, we show the location of the eight stations used for taking measurements. We used a Google map to locate the places of each station. Nature of Data: A complete data set of the meteorological factors and air pollution parameters PM10, with particulate matter of a diameter less than 10µm and TSP, with particles larger than 10µm in diameter, was carried out in eight monitoring stations around Al-Fuhais cement plant over the one-year period.The collected data during the study period are from 26 November 2006 to 25 November 2007. Ambient air quality was monitored with a sampling frequency of 24 hours. The stations were distributed to cover different types of environments in the area. They were distributed around the main industrial activities. The meteorological data include Temp, RH, WS were obtained from Al-Salt meteorological station (the nearest station to the study area). In Figure 5 and 6, we show the training and testing data used to develop the ANNARX models for both the PM10 and the TSP air pollution parameters. Ann Model Building: The most general structure for nonlinear black-box modeling technique is a neural input-output model, which can be treated as a nonlinear extension of linear dynamic systems. Using ANN in the

Fig. 3: High volume sampler GS3210

Fig. 4: Locations of air quality monitoring stations at Al-Fuhais City, Salt, Jordan identification of nonlinear dynamic systems has been growing in recent years and is recommended for cases where there is no a priori knowledge about the real system; also it is useful when the system to be modeled has a very complex nonlinear dynamic behavior [33, 34]. ANNs have the capability to cover a nonlinear mapping from past signals (past output signals and/or past input signals) to the predicted value of y(t). Authors [39, 40] have presented many designs, architectures and learningalgorithms to deal with different types of system identification using ANNs [41]. General nonlinear input-output models can be described as in Equation 6 [40]: 1003


Fig. 5: Training data set for both the PM10 and TSP ANN Models

Fig. 8: The proposed ANNs Architecture for the PM10 ANN model y (t ) yˆ (t | ) + e(t ) =

y (t ) g [ (t | ), =

] + e(t )

(4)

where (t, ) is the regression vector, is the neural network adjustable parameters vector known as the weights W and g is a nonlinear function realized by the neural network. The proposed ANNARX Input Model is presented in Figure 7. The regression vector [42] is defined as in Equation 7. (t )=

[ y (t − 1),..., y (t − nA), u(t − k ),..., u (t − k − nB )]T

(7)

Notice that the regression vector is formed with the past values of the input and output of the system. Fig. 6: Testing data set for both the PM10 and TSP ANN Models

Proposed ANN PM10 and ANN TSP Models: The proposed ANN architecture is given as three layers as follows (Figure 8 and Figure 9): The input layer has seven input neurons. The hidden layer has three neurons for the PM10 and five for the TSP with nonlinear sigmoid function. The hidden units are fully connected to both the input and output. The output layer has one linear output neuron to produce the prediction of PM10/TSP computed at ”Deir Al-Latin School”.

Fig. 7. The NNARX Model Structure [42]

Model Evaluation: In order to check the performance of the developed ANN models the Mean Squares Error (MSE), the Euclidian distance (ED), the Manhattan 1004


Mean Squares Error (MSE) = MSE

1 n

n

∑ ( yi − yî )2 i =1

(8)

Euclidian distance (ED): n

∑ ( yi − yî )2

= ED

(9)

i =1

Manhattan distance (MD): = MD (

n

∑ yi − yî ) i =1

Mean Magnitude of Relative Error (MMRE)

Fig. 9: The proposed ANNs Architecture for the TSP ANN model

n yi − yî 1 ) MMRE = ( n i =1 yi

∑

Table II: Inputs and Outputs in Both the Pm10 and Tsp Ann Models Inputs

Temp(t-1), Temp(t-2), RH(t-1), RH(t-2) WS(t-1), WS(t-2), y(t-1)

Output

y(t)

Table III: Setting Parameters for Adopted Ann Models Parameters

Value

Number of layers

3

Number of neurons in the hidden layer for the PM10 ANN Model

3

Number of neurons in the hidden layer for the TSP ANN Model

5

Number of neurons in the output layer

1

Number of iterations

500

Table IV: Evaluation Criteion for Pm10 Dataset Criteria

Training

Testing

MSE

155.8054

219.7853

MD

9.3084

11.2080

ED

192.5660

161.0424

MMRE

0.3133

0.3125

Table V: Evaluation Criteion for Tsp Dataset Criteria

Training

Testing

MSE

506.0165

1010.7

MD

16.8033

23.6472

ED

347.0330

345.3467

MMRE

0.2902

0.2344

distance (MD) and the MeanMagnitude of Relative Error (MMRE) were measured. Theseperformance criterions are assessed to measure how close themeasured values to the values developed using the ANNapproach. These criterions helps in better evaluating thecapabilities of the developed ANNARX models for thePM10 and TSP. MSE, ED, MD and MMRE are computedas follows:

(10)

(11)

wherey and yˆ are the observed PM10 or TSP values and the predicted values based on proposed model and n is the number of measurements used in the experiments, respectively. Experimental Results: To develop the ANNARX models, we usedthe NNSYSID MATLAB Toolbox. This toolbox was providedby Magnus Norgaard in [43]. The MATLAB Toolbox consists of a set of tools for neural network based identification of nonlinear dynamic systems. It has many ANN learning algorithms implementation. We adopted the ANNARX model which is presented in Figure 8 and Figure 9 with the appropriate input delays. Given that y(t) is considered the ANN predicted output; which is either PM10 or TSP according to our case study. The BP learning algorithm was used for training the ANNARX to satisfy the minimum error convergence requirements. The prediction error method is based on the introduction of a measure of closeness specified in terms of the Mean Square Error (MSE) error criteria. ANN Setup: We adopted the given architecture in Figure 8 and Figure 9 along with the setup of the ANN parameters given in Table III. PM10 ANN Model: We run the proposed ANNARX model for number of experiments. The results were collected after each experiment to select the best ANN performance model for predicting the PM10. It was found 1005


Fig. 10: Observed and predicted measurements for the PM10 ANN Model in the Training Case

Fig. 13: Observed and predicted measurements for the TSP ANN Model in the Training Case

Fig. 11: Observed and predicted measurements for the PM10 ANN Model in the Testing Case

Fig. 14: Observed and predicted measurements for the TSP ANN Model in the Testing Case that the developed ANN is quite promising and was capable of achieving acceptable error level after number of iterations using three neurons in the hidden layer. Figure 10 illustrates the performance of the model over the training data set. On the other hand, Figure 11 illustrates the performance of the model over the testing data set. The computed difference between the observed and predicted PM10 is presented in Table IV. ANN convergence during the training process is shown in Figure 12 over five experiments.

Fig. 12: MSE Convergence for PM10 ANNs Model

TSP ANN Model Results: We found that the relationship between the inputs and output variables of the TSP model is more complex than in the case of PM10. That is why we introduced more neurons in the 1006


REFERENCES 1.

Fig. 15: MSE Convergence for TSP ANNs Model hidden layer for the ANN (five neurons). We run ANN over five experiments and collected the results in each case. The best produced results are shown in Figure 15. Figure 15 shows the convergence of the proposed ANN TSP model over five experiments. The observed and predicted TSP values in both training and testing cases is shown in Figures 13 and 14. In Table V, we show the performance results obtained based the ANN TSP model. CONCLUSIONS In this paper we explored the use of ANN to build a nonlinear ARX model for both the PM10 and TSP airpollution parameters. Although we used a short period of time to develop our models (one year) the developed models seems to be quite good with respect to adopted error evaluation criterion. We believe that we can provide better modeling results if we explored other learning algorithms which can provide a more robust results. We plan also to explore the use of hybrid learning algorithms such as Artificial Neural Fuzzy Inference Systems (ANFIS) for process learning. The learning algorithm showed tremendous modeling capabilities in various application areas. ACKNOWLEDGEMENTS Authors would like to thank the Faculty of Agricultural Technology, Al-Balqa Applied University (BAU) for providing the necessary data collected at eight monitoring stations to measure PM10 and TSP concentrations at Al-Fuhais, Salt, Jordan.

Leksmono, N.S., J.W.S. Longhurst, K.A. Ling, T.J. Chatterton, B.E.A. Fisher and J.G. Irwin, 2006. Assessment of the relationship between industrial and traffic sources contributing to air quality objective exceedences: a theoretical modelling exercise, Environ. Model.Softw., 21: 494-500. 2. Mallet, V. and B. Sportisse, 2008. Air quality modeling: From deterministic to stochastic approaches, Comput. Math. Appl., 55: 2329- 2337. 3. Bert Brunekreef, S. and S.T. Holgate, 2002. Air pollution and health, THE LANCET, 360: 1233-1242. 4. Bai, H., Y. Ma, X. Ai, H. Li, P. Liu and D. Cang, 2011. Chemical and morphological properties of particulate matter generated from coal fired circulating fluidized bed boiler, in Proceedings of the 2011 Third International Conference on Measuring Technology and Mechatronics Automation (ICMTMA ’11), Volume 01, (Washington, DC, USA), pp: 708-711, IEEE Computer Society. 5. Carmichael, G.R., A. Sandu, T. Chai, D.N. Daescu, E.M. Constan- tinescu and Y. Tang, 2008. Predicting air quality: Improvements through advanced methods to integrate models and measurements, J. Comput. Phys., 227: 3540-3571. 6. Sokhi, R.S., H. Mao, S.T.G. Srimath, S. Fan, N. Kitwiroon, L. Luhana, J. Kukkonen, M. Haakana, A. Karppinen, K. Dick van den Hout, P. Boulter, I.S. McCrae, S. Larssen, K.I. Gjerstad, R. San Jose´, J. Bartzis, P. Neofytou, P. van den Breemer, S. Neville, A. Kousa, B.M. Cortes and I. Myrtveit, 2008. An integrated multi-model approach for air quality assessment: Development and evaluation of the oscar air quality assessment system, Environ. Model.Softw., 23: 268-281. 7. Finardi, S., R. De Maria, A. D’Allura, C. Cascone, G. Calori and F. Lollobrigida, 2008. A deterministic air quality forecasting system for torino urban area, italy,” Environ. Model.Softw., 23: 344-355. 8. Appel, K.W., R.C. Gilliam, N. Davis, A. Zubrow and S.C. Howard, 2011. Overview of the atmospheric model evaluation tool (AMET) v1.1 for evaluating meteorological and air quality models, Environ. Model.Softw., 26: 434-443. 9. Zolghadri, A. and F. Cazaurang, 2006. Adaptive nonlinear state-space modelling for the prediction of daily mean PM10 concentrations, Environ. Model.Softw., 21: 885-894. 10. Nunnari, G. and F. Cannavo´, 2007. A new cost function for air quality modeling, J. VLSI Signal Process. Syst., 49: 281-290.

1007


11. Oprea, M. and A. Matei, 2010. The neural networkbased forecasting in environmental systems, WSEAS Trans. Sys.Ctrl., 5: 893-901. 12. Oprea, M. and A. Matei, 2010. Applying artificial neural networks in environmental prediction systems,” in Proceedings of the 11th WSEAS International Conference on Automation & Information (ICAI’10), (Stevens Point, Wisconsin, USA), pp: 110-115, World Scientific and Engineering Academy and Society. 13. Neagu, C.D., N. Avouris, E. Kalapanidas and V. Palade, 2002. Neural and neuro-fuzzy integration in a knowledge-based system for air quality prediction, Applied Intelligence, 17: 141-169. 14. Kalapanidas, E. and N. Avouris, 2003. Feature selection for air quality forecasting: a genetic algorithm approach, AI Commun., 16: 235-251. 15. Mensink, C., F. Deutsch, L. Janssen, R. Torfs and J. Vankerkom, 2007. Comprehensive modelling of PM10 and PM2.5 scenarios for Belgium and Europe in 2010, in Proceedings of the 6th international conference on Numerical methods and applications (NMA’06), (Berlin,Heidelberg), pp: 466-474, SpringerVerlag, 2007. 16. Kunzli, N., R. Kaiser, S. Medina, M. Studnicka, O. Chanel, P. Filliger, M. Herry, F.J. Horak, V. Puybonnieux-Texier and P. Queńel, 2000. Public- health impact of outdoor and traffic-related air pollution: a european assessment,” THE LANCET, 356: 795-801. 17. Mn, R. and R. Hvn, 2011. Air Pollution. Tata Mcgraw Hill Publishing Co Ltd. 18. Amann, M., I. Bertok, J. Borken-Kleefeld, J. Cofala, C. Heyes, L. Hoglund-Isaksson, Z. Klimont, B. Nguyen, M. Posch, P. Rafaj, R. Sandler, W. Scho¨pp, F. Wagner and W. Winiwarter, 2011. Cost-effective control of air quality and greenhouse gases in europe: Modeling and policy applications, Environ. Model.Softw., 26: 1489-1501. 19. Slini, T., A. Kaprara, K. Karatzas and N. Moussiopoulos, 2006. PM10 forecasting for thessaloniki, greece, Environ. Model. Softw., 21: 559-565. 20. Silibello, C., G. Calori, G. Brusasca, A. Giudici, E. Angelino, G. Fossati, E. Peroni and E. Buganza, 2008. Modelling of PM10 concentrations over Milano urban area using two aerosol modules, Environ. Model. Softw., 23: 333-343.

21. Liu, X., Y. Liang, W. Zhou, X. Peng, X. Tang and Y. Wang, 2011. Pollution characteristics of heavy metal in the PM10 of Nanchang City, in Proceedings of the 2011 Second International Conference on Digital Manufacturing & Automation, ICDMA ’11, (Washington, DC, USA), pp: 859-862, IEEE Computer Society, 2011. 22. Y. Zhou, Y. Li, G. Huang, W. Zhang and Y. Zhou, 2006. Planning energy and environmental systems associated with air pollutants mitigation under uncertainty, in Proceedings of the 2012 International Con- ference on Biomedical Engineering and Biotechnology, ICBEB ’12, (Washington, DC, USA), pp: 1594-1597, IEEE Computer Society, 2012. [23] C. Carnevale, V. Gabusi and M. Volta, “Poem-pm: an emission model for secondary pollution control scenarios, Environ. Model.Softw., 21: 320-329. 24. Hamasha, K.M., M.S. Almomani, M. Abu-Allaban and W.P. Arnott, 2010. Study of black carbon levels in city centers and industrial centers in Jordan, Jordan Journal of Physics, 3(1): 1-8. 25. Al-Hasaan, A.A., 1989. Air Pollution Monitoring in Amman: A Collaborative Project. Royal Scientific Society of Jordan Report. 26. Al-Hasaan, A.A., T.F. Dann and P.F. Brunet, 1992. Air pollution monitoring in amman, jordan, J. Air Waste Manage. Assoc., 11: 814-816. 27. Hussein, T., R.A. Al-Ruz, T. Petaja, H. Junninen, D.E. Arafah, K. Hameri and M. Kulmala, 2011. Local air pollution versus shortrange transported dust episodes: A comparative study for submicron particle number concentration, Aerosol and Air Quality Research, 11: 109119. 28. Vemuri,V., 1988. Artificial neural networks: theoretical concepts, ch. Artificial neural networks: an introduction, pp: 1-12, Los Alamitos, CA, USA: IEEE Computer Society Press. 29. Braspenning, P.J., F. Thuijsman and A.J.M.M. Weijters, eds., 1995. Artificial Neural Networks: An Introduction to ANN Theory and Practice, (London, UK, UK), Springer-Verlag. 30. Duch, W., J. Kacprzyk, S. Zadrozÿny and E. Oja, eds., 2005. Proceedings of the 15th international conference on Artificial Neural Networks (ICANN’05): formal models and their applications - Volume Part II, (Berlin, Heidelberg), Springer-Verlag.

1008


31. Olawoyin, R., A. Nieto, R.L. Grayson, F. Hardisty and S. Oyewole, 2013. Application of artificial neural network (ann)-self-organizing map (som) for the categorization of water, soil and sediment quality in petrochemical regions, Expert Syst. Appl., 40: 3634-3648. 32. Baareh, A., A. Sheta and K.A. Khnaifes, 2006. Forecasting river flow in the USA: A comparison between auto-regression and neural networknonparametric models, Journal of Computer Science, USA, 2: 10. 33. Eghnam, K. and A. Sheta, 2008. Solving capelin time series ecosystem problem using hybrid ANN-GAs model and multiple linear regression model, in Proceedings of the 1st Mediterranean Conference on Intelligent Systems and Automation, edited by H. Arioui, R. Merzouki and H.A. Abbassi, 30(1): 482-486. 34. H. Al-Hiary, A. Sheta and A. Ayesh, 2008. Identification of a chemical process reactor using soft computing techniques, in Proceedings of the 2008 International Conference on Fuzzy Systems (FUZZ2008) within the 2008 IEEE World Congress on Computational Intelligence (WCCI2008), Hong Kong, 1-6: 845-653. 35. Sheta, A., H. Al-Hiary and M. Braik, 2009. Identification and model predictive controller design of the Tennessee Eastman chemical process using ANN, in Proceedings of the 2009 International Conference on Artificial Intelligence (ICAI’09), July 13-16, 2009, USA, 1: 25-31. 36. Ömer Faruk, D., 2010. A hybrid neural network and ARIMA model for water quality time series prediction, Eng. Appl. Artif. Intell., 23: 586-594.

37. Radakovic, B. and T. Krope, 2007. The effect of ambient air pollution on human health, in Proceedings of the 2nd IASME/WSEAS international conference on Energy and environment, EE’07, (Stevens Point, Wisconsin, USA), pp: 253-256, World Scientific and Engineering Academy and Society (WSEAS), 2007. 38. Koop, G., R. McKitrick and L. Tole, 2010. Air pollution, economic activity and respiratory illness: Evidence from canadian cities, 1974-1994, Environ. Model.Softw., 25: 873-885. 39. Narendra, K.S. and K. Parthasarathy, 1991. Gradient methods for the optimization of dynamical systems containing neural networks, IEEE Trans. on Neural Networks, pp: 252-262. 40. Norgaard, M., O. Ravn and N.K. Poulsen, 2001. NNSYSID and NNCTRL tools for system identification and control with neural networks, Computing & Control Engineering Journal, 12(1): 29-36. 41. Czop, P., 2005. System Identification in Fault Detection and Isolation. PhD thesis, University of Hsinchu, Taiwan, ROC. 42. Norgaard, M., O. Ravn, N.K. Poulsen and L.K. Hansen, Neural Networks for Modelling and Control of Dynamic Systems.Springer, London. 43. Norgaard, M., O. Ravn and N. Poulsen, 2002. Nnsysid - toolbox for system identification with neural networks, Mathematical and computer modelling of dynamical systems, 8(1): 1-20.

1009