Application of neural network of Multi Layers Perceptron (MLP) in site selection of waste disposal (Case ‎study: fereydoonshahr city)‎

Introductionþ:þCities are at the nexus of a further threat to the environment, namely the production of an increasing ýquantity and complexity of wastes. The estimated quantity of Municipal Solid Waste (MSW) generated ýworldwide is 1.7 – 1.9 billion metric tons.ý
ý In many cases, municipal wastes are not well managed in low-income countries, More than 50 per cent of ýthe collected waste is often disposed of through uncontrolled landfilling and about 15 per cent is ýprocessed through unsafe and informal recycling. Municipal Solid Waste (MSW) is the natural result of ýhuman activities. If an appropriate management system is not used for this problem, it may lead to ýenvironmental pollution and jeopardize the mankind’s health. ý
The ANN models are basically based on the perceived work of the human brainþ.þý ANNs can be trained to ýmodel any relationship between a series of independent and dependent variables (inputs and outputs to ýthe network respectively). For this reason, ANNs have been usefully applied to a wide variety of ýproblems that are difficult to understand, define, and quantify. It should be pointed out that similar to any ýother statistical and mathematical model, ANN models have also some disadvantages, too. Having a large ýnumber of input variables is one of the most common problems for their development because they are ýnot engineered to eliminate superfluous inputs.ý
ý Literature survey demonstrates that artificial neural network (ANN) models are proper tools for prediction ýof solid waste generation predicting. Noori et al (2008) investigated the Prediction of Municipal Solid ýWaste Generation with Combination of Support Vector Machine and Principal Component Analysis in ýMashhad and in authors’ opinion, the model presented in this article is a potential tool for predicting WG ýand has advantages over the traditional SVM model. Jalili and Noori (2008) investigated the Prediction of ýMunicipal Solid Waste Generation by Use of Artificial Neural Network and Results point that artificial ýneural network model has more advantages in comparison with traditional methods in predicting the ýmunicipal solid waste generation.Noori et al (2010) investigated the Evaluttion of PCA and Gamma test ýtechniques on ANN operation for weekly solid waste prediction and Findings indicated that the PCA-ANN ýand GT-ANN models have more effective results than the ANN model. These two models decrease the ýnumber of input variables from 13 to 7 and 5, respectively.ý
The accurate prediction of waste disposal Zonation plays an important role in the solid waste management ýsystem. For this reason, ANN is used and different models are created and tested.ý
Materialsþ & þý Methodsþ:þFereydunshahr city is located from 49° 36ʹ to 50° 19ʹ longitude and from 32° 37ʹ to 33° 05ʹ latitude ýgeographic coordinate system. The extent of the area is 77646 hectar. Fereydunshahr city with an average ýaltitude of 2500 m above sea level is a mountainous region and is located in the province of Isfahan. ýAccording to hydrological, geological, and Geomorphological characteristics of study area and the goals ýoutlined, it can be said that the parameters used to Municipal Solid Waste landfilling are different. In this ýresearch the most important factors are used For this purpose are 12 primary factors influencing ýMunicipal Solid Waste landfilling in the study area, including lithology, Level of groundwater, Soil ýtexture, distance to habitate, land use, slope, aspect, elevation, rainfall, distance to fault, distance to road, ýand drainage density were identified by interpretation of satellite imagery, aerial photography, and field ýstudies. The used base map in this work including geological map at a scale of 1: 100,000, aerial ýphotographs on a scale of 1: 40,000, topographical maps with a scale of 1: 50,000, ETM 놫庖墭 images ýand precipitation (rain-gauge stations) were prepared by ArcGIS10.2 software. ý
The digital elevation model (DEM) with 30 meter multiplied by 30 meter pixel size was prepared by using ýtopographic map 1:50000. The distance to drainage and road was extracted by drainage and road ýnetworks from study area topographic map. The land use map was provided by including unsupervised ýclassification ETM image satellite, field survey, and accuracy control. Also geologic map was prepared ýby digitizing and polygonize of rock units of geologic map 1:100000 and using ArcGIS10.2. Artificial ýneural networks, originally developed to mimic basic biological neural systemsþ.þý a network can perform a ýsurprising number of tasks quite efficiently (Reilly and Cooper,1990). This information processing ýcharacteristic makes ANNs a powerful computational device and able to learn from examples and then to ýgeneralize to examples never before seen. Recent research activities in artificial neural networks (ANNs) ýhave shown that ANNs have powerful pattern classification and pattern recognition capabilities.The most ýpopular architecture for a neural network is a multilayer perceptron (Bishop, 1995; Jain, et al., 2006). In ýthis study, we used was the feed forward, multilayer perceptron (MLP), which is consideredable to ýapproximate every measurable function (Gardner and Dorling, 1998). The main issue in training MLP for ýprediction is the generalization performance. MLP, like other flexible nonlinear estimation methods such ýas kernel regression, smoothing splines, can suffer from either underfitting or overfitting (Coulibaly, et al., ýý2000). In this situation error between training and testing results start to increase. For solving this problem, ýStop Training Approach (STA) has been used. Data are divided into 3 parts in this method. First part is ýrelated to network training, second part for stopping calculations when error of integrity start to increase ýand the third part that is used for integrity of network. In order to evaluate the performance of the ANN ýmodel 3 statistical indices are used: t Mean Squared Normalized Error (MSE)ý, root mean square error ýý(RMSE) and correlation coefficient (R2) values that are derived in statistical calculation of observation in ýmodel output predictions, defined as:ý MSE=(∑_(i= 0)^N▒〖 (d_(i )- y_i ) 〗)/N ý RMSE=√(∑_(i=1)^n▒((obs-pre)/n) ) ýþ þ R^2=(∑_(i=1)^n▒〖(obs-obs) (pre-pre) 〗)/(√(∑_(i=1)^n▒(obs-obs)^2 ) ∑_(i=ýý1)^n (pre-pre)^2 )ýþ þ
Discussion of Resultsþ þ& ýAccurate prediction of landfilling site selection of municipal solid waste is crucial for programming ýmunicipal solid waste management system. In this research with application of feed forward artificial ýneural network, an appropriate model for predicting of landfilling site selection of municipal solid waste ýin Fereydunshahr city, was proposed. For this purpose, In this paper, neural network is trained and tested ýusing MATLAB 7.2.. For this purpose, 12 primary factors influencing Municipal Solid Waste landfilling in ýthe study area, including lithology, Level of groundwater, Soil texture, distance to habitate, land use, ýslope, aspect, elevation, rainfall, distance to fault, distance to road, and drainage density were chosen for ýimput layers. Also, for recognizing the effect of each input data sensitive analysis was performed.Finally, ýdifferent structures of artificial network were investigated and then the best model for predicting ýlandfilling site selection of municipal solid waste was chosen based on ýMean Squared Normalized Error ýý(MSE)ý, root mean square error (RMSE) and correlation coefficient (R2) indexes. After performing of the ýmentioned models, Mean Squared Normalized Error (MSE)ý, root mean square error (RMSE) and ýcorrelation coefficient (R2)in neural network for test have been achieved equal to 0.0081 , 0.11 and ýý0.999% respectively. Results indicate that trainlm model has more advantages in comparison with trainbp ýand trainbpx methods in landfilling site selection of municipal solid waste. after determining the best ýnetwork structure, zonation map of the best site for landfilling of municipal solid waste using 12 imput-ýlayer was prepared in 5 classes. The results showed that 37.2% (28884/31Ha) of the total area is very ýsuitable for waste landfilling, 7.2% (5590/51 Ha) suitable, 12.6% (9783/39 Ha) is fairly suitable, 38% ýý(29505/48 Ha) unsuitable and 5% (3882/3 Ha) is very unsuitableý.