Expert Systems with Applications

Expert Systems with Applications 38 (2011) 11999–12008 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: ...
Author: Matthew Lang
4 downloads 1 Views 1MB Size
Expert Systems with Applications 38 (2011) 11999–12008

Contents lists available at ScienceDirect

Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa

Real world representation of a road network for route planning in GIS Abolghasem Sadeghi-Niaraki a,⇑, Masood Varshosaz b, Kyehyun Kim a, Jason J. Jung c a

Dept. of Geoinformatic Eng., Inha Univ., Incheon, South Korea Dept. of Geodesy and Geomantic Eng., K.N. Toosi Univ. of Tech., Tehran, Iran c Dept. of Computer Eng., Yeungnam Univ., Gyeongsan, South Korea b

a r t i c l e Keywords: Route planning Sensitivity analysis GIS AHP Road network Impedance model

i n f o

a b s t r a c t This paper addresses a methodology to properly represent a road network in the geographic information system (GIS) for network analysis. Over the years, the real world has become too complex to model properly within a given information system, such as GIS. Ideally, when the real world is represented as accurately as possible, a GIS can answer a question in its virtual world that coincides with the exact answer in the real world. However, existing methods related to impedance modeling for each segment of a road network in a route planning analysis that includes only a distance or time variable do not give proper results. Hence, this study investigates how a road network can represent the real world in a GIS and offer route planning tools. To address this, first, additional realistic variables are taken into account. These include weather, sight-seeing information, road type, and so on. Second, to combine these variables, an impedance model (IM) using the analytical hierarchical process (AHP) method is proposed. Finally, all of the models are implemented and verified with a sensitivity analysis. The models were successfully implemented in this work. All of the paths of the route planning analysis were successfully matched with the drivers’ paths that would normally be chosen in reality. It is anticipated that the use of other techniques such as analytical network process (ANP) in addition to AHP would be useful to overcome the aforementioned problem. Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction A significant issue in the field of object modeling is proper representation of objects in the real world inside a geographical information system (GIS) environment. Better representation of the real world can lead to proper utilization of GIS tools. Network analysis, a kind of spatial analysis, is a powerful tool in the GIS environment for calculating the optimum path in a network. In general, a network is a system of interconnected linear features through which resources are transported or communication is achieved. A network data model is an abstract representation of the components and characteristics of real-world network systems. A network model can be defined as a line graph, which is composed of links representing linear channels of flow and nodes representing their connections (Lupien, Moorl, & Dagermond, 1987). In other words, a network takes the form of edges (or arcs) connecting pairs of nodes (or vertices). Nodes can be junctions and edges can be segments of a road or a pipeline. For a network to function as a realworld model, an edge must be associated with a direction and with a measure of impedance or cost, determining the resistance or travel cost along the network (Husdal, 2000a). ⇑ Corresponding author. Tel.: +82 32 860 7600; fax: +82 32 863 1506. E-mail address: [email protected] (A. Sadeghi-Niaraki). 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.12.123

One major application of network analysis is found in transportation planning, where the goal might be to find paths corresponding to certain variables, such as finding the shortest or least cost path between two or more locations, or to find all locations within a given travel cost from a specified origin (Husdal, 2000a). Traditionally, network analysis, path finding, and route planning have been in the domain of graph theory and vector GIS, the areas where most path finding algorithms find their application (Husdal, 2000b). Dolan and Aldous (1993), Chou (1997), and Jones (1998) have described the process of finding a variables-determined path through a network in great detail. In a network analysis the ‘‘cost’’ or ‘‘impedance’’ of the individual segment of network is frequently used as a mono-dimensional variable such as distance in Chunithipaisan, James, and Parker (2004), speed in Leonard et al. (2000), traffic in Shadewald, Hallmark, and Souleyrette (2001), and so on. There is a significant problem regarding network analysis in most GIS environments. For instance, the use of a mono-dimensional variable as an impedance of each segment yields unrealistic results when finding a preferred path between the origin and destination in a network. Although some studies reported several variables other than mono variables for impedance of each segment, but they did not model both quantitative and qualitative multi-dimensional parameters and the approaches did not provide a general methodology for

12000

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

any case. Furthermore, most of those studies focused on roads inside of cities, not inter-city roads. For example, Jun, Koo, and Koh (2004) proposed the use of regional supply, distance, road network, and construction cost. Thirumalaivasan and Guruswamy (1997) noted that applying various impedance factors that play a significant role in deciding the travel time such as volume of traffic, type of road, road width, number of junctions, turns etc. Their method were based on empirical formula not any generic mathematical method. The aforementioned problems show that a GIS network based on a mono-dimensional impedance variable such as ‘‘distance’’ cannot properly represent the situation of real world. Hence, any network analysis process in a network based on mono-dimensional variables leads to unrealistic results. In order to arrive at suitable and realistic results in a GIS environment when using a route finding algorithm (like Dijkstra, 1959) it is necessary to represent or simulate properly the real world via a methodology that can determine the exact specifications of the road network in a GIS environment and thereby deliver user satisfaction. Therefore, for representing and simulating the situation of real world road in GIS, realizing several new and efficient parameters beyond merely the distance or time for each segment is required. This research employs a concept that can be easily implemented for any kind of network but focuses on a road network. The rationale for this is that roads constitute one of the most popular networks in GIS environments, as compared with other networks such as gas and river networks. Furthermore, in the real world there are several situations for road networks that GIS normally does not address, such as extent of road traffic, weather, scenery, facilities around the road, and so on. In any route finding algorithm based on the lowest amount of impedance between the origin and destination, the final path among several paths between those points is selected. One long path between the start point and endpoint in the network could be equal to many small segments in the network. In this regard, the sums of the impedances of the individual segments of the network are used to estimate the total impedance among all possible paths between the origin and destination in order to select the optimum or shortest path. Therefore, it is noted here that distance alone is not sufficient to show all situations around each segment of the road network. The objectives of this study are to derive appropriate and related variables that affect each road segment during the network analysis and to merge these variables to develop an appropriate impedance model in GIS for route planning using AHP techniques. The impedance model yields route planning results that are very close to reality from a given origin to destination in the road network.

of the accessible road variables that will influence the road in implementing the impedance model. The road variables include factors that affect individual road segments to define the impedance of the road segments when necessary for route planning. These variables include weather conditions, tourism (sightseeing attractions), road traffic, security, facilities, and length. For these variables several sub-variables were also selected. The weather conditions include moderate, relatively dry, cold, desert, warm and relatively dry, and warm and humid weather conditions. The sub-variables of tourism consist of the effects of the sea, mountain regions, recreational regions, culturalhistorical sites, antiquities-related sites, and religious sites, as well as dikes, forest zones, rivers or streams, ski resorts, and lakes. For road traffic, several levels of service (A, B, C, D, and E) were selected as sub-variables. The existence of police offices, side-road parking lots, car service centers, health and medical treatment services, and telephone booths were obtained as sub-variables for the variable security in this research. Sub-variables of facilities include fuel stations and public service centers, and terminals. In this research the AHP enables a hierarchical formulation and allows different combinations of various factors in the model. AHP, developed by Thomas L. Saaty in 1975, is based on a method of analyzing complicated fuzzy matters using the human brain. Many applications of this method have been suggested by different researchers (Qodsipoor, 1999). In the AHP process, the first step is decomposition, or the structuring of the problem into a hierarchy. This hierarchic structuring reflects the natural tendency of the mind to sort elements of a system into different levels and to group like elements in each level (Stewart, 2005). In this study, the AHP flowchart addresses two levels. The elements of the first level are the aforementioned main variables (weather conditions, tourism (sightseeing attractions), road traffic, security, facilities, and length.) and those of the second level are composed of the sub-variables of each main variable noted above. A comparative judgment matrix is the next step of the AHP process. The elements on the first level are arranged into a matrix and the decision maker makes judgments regarding the relative importance of the elements with respect to the overall goal (Saaty & Vargas, 1991). The result of AHP modeling in the first level leads to the impedance model shown in Eq. (1). In this model, the weights of the variables have been derived from AHP pairwise comparisons. In fact, the modeling of this research has the capability to define more scenarios based on various variables on several different conditions. Thus, in the process of defining these scenarios, the research covered all possible and important conditions that have high a probability to occur for this trip.

IM ¼

n X

ð1=Di ð0:149T 1 þ 0:296T 2 þ 0:193T 3 þ 0:175T 4 þ 0:187T 5 ÞÞ

i¼1

2. Impedance modeling In this research the first step of impedance modeling is the selection of variables. All of the variables that represent all the characteristics of the road network in the real world are divided into three groups, which are respectively comprised of road, vehicle, and human factors. These variables were selected based on their influence on route planning in the road network. The road traffic, security of the road, the weather conditions, and the road type are some examples of road variables. The vehicle variables and the extent of the fuel capacity are examples of the second group. The last group is dedicated to human factors including the age and skill of the drivers, as well as the capacity of the drivers to understand the technical knowledge of the automobiles. Among the above three factors, the road factor is more usable and accessible in terms of route planning. Therefore, this study uses only some

ð1Þ where IM states the impedance model (IM) of each road network, T1 presents the weather conditions around the road network, T2 states the number of tourist places that affect the road network, T3 is the extent of the road traffic flow variable, T4 expresses the number of security (safety) points around the road network, T5 presents the number of facilities around the road network, and Di is the length of an individual road segment. The length variable of each segment (Di) has a special attitude compared with other variables; i.e. it has a reverse relationship with other variables. Furthermore, the coefficients of this equation are normalized, implying that the aggregate of those coefficients is equal to 1. To express each variable of Eq. (1) in detail, the following section provides further explanations of these variables. Each main variable of this equation includes various sub-variables. The following equations were derived after AHP processing for each related sub-variable.

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

12001

The weather variable (T1) includes six sub-variables. After carrying out the AHP process for weather sub-variables, Eq. (2) is obtained.

so that the volume carried by the roadway falls below its capacity; without the stoppages, the volume of traffic on the roadway would be higher, or in other words, it would reach capacity (Dush & Muhonen, 2002). Eq. (4) shows the result of the AHP process for road traffic with respect to LOS

T 1 ¼ ð0:378X 1 þ 0:275X 2 þ 0:128X 3 þ 0:096X 4 þ 0:079X 5 þ 0:045X 6 Þ

T 3 ¼ ð0:505Z 1 þ 0:267Z 2 þ 0:127Z 3 þ 0:074Z 4 þ 0:027Z 5 Þ

2.1. The weather variable

ð2Þ where T1 is the weather variable that affects the road segment, X1 represents a cold area around the road network, X2 denotes moderate weather, X3 represents a dry-cold weather around the road network, X4 expresses warm-humid weather, X5 presents dry-warm conditions around the road, and X6 is desert areas around road segments. 2.2. The tourism variable The surroundings of the road network normally include very popular sight-seeing areas. Thus, the tourism variable is significant in tourist trip impedance modeling. After performing the AHP method only for tourism variables in level two, Eq. (3) is derived for the tourism variables

T 2 ¼ ð0:209Y 1 þ 0:169Y 2 þ 0:197Y 3 þ 0:045Y 4 þ 0:023Y 5 þ 0:170Y 6 þ 0:057Y 7 þ 0:114Y 8 þ 0:016Y 9 Þ

ð3Þ

where T2 is the tourist variable, Y1, Y2, Y3, Y4, Y5, Y6, Y7, Y8, and Y9 present the effect of the sea, mountain regions, the recreational regions, places (cultural-historical, antique-related, and religious places), dikes, forest zones, rivers or streams, ski resorts, and lakes, respectively. 2.3. The road traffic variable The traffic variable is a significant variable in the impedance modeling process. Normally, two types of methods are used to consider the traffic extent. The first is the use of the average daily traffic (ADT) or annual average daily traffic (AADT) data. The second is the application of the level of service (LOS) variable (Behbahani, 1997). Calculating the ADT and AADT entails problems such as consideration of traffic without regard to the type of road, the volumeto-capacity ratios, and the type of area in question. Thus, in this work, the LOS variable was utilized. The concept of LOS is also described in the highway capacity manual (HCM, 1994). It involves qualitative measures that characterize operational conditions within a traffic stream and their perception by motorists and passengers (Yamzon, 2005). Roadway LOS is a measure of roadway congestion ranging from LOS A (least congested) to LOS F (most congested). LOS is one of the most common terms used to describe how ’’good’’ or how ‘‘bad’’ traffic is projected to be. There are six levels of service letter grades typically recognized by transportation planners and engineers. They are as follows: LOS A describes a condition of free flow, with low volumes and high speeds. LOS B is the zone of stable flow, with operating speeds beginning to be restricted somewhat by traffic conditions. Nonetheless, drivers in n LOS B zones have reasonable freedom to select their speed and lane of operation. LOS C has mostly stable flow, but speeds and maneuverability are more closely constricted by the higher volumes. LOS D is a zone that approaches an unstable flow, with tolerable operating speeds. Driving speed, however, is considerably affected by changes in the operating conditions. LOS E is a zone that cannot be described by speed alone. Operating speeds are lower than in Level D, with volume at or near the capacity of the highway. LOS F is a zone in which the operating speeds are controlled by stop-and-go mechanisms, such as traffic lights. This is known as a forced flow operation. The stoppages disrupt the traffic flow

ð4Þ

where V3 states the road traffic flow, and Z1, Z2, Z3, Z4, and Z5 present the LOS A, LOS B, LOS C, LOS D, and LOS E of that segment of road network, respectively. 2.4. The security variable In this research, the security variable acts as a sub-model with several sub-variables that can take into account the surrounding situation of the road, as derived in Eq. (5)

V 4 ¼ ð0:088N1 þ 0:088N2 þ 0:033N3 þ 0:406N4 þ 0:298N5 þ 0:069N 6 þ 0:017N 7 Þ

ð5Þ

where V4 denotes the security around the road network, N1 gives the number of police station offices that control that segment of road network, N2 states the number of road maintenance offices entrusted with road responsibility from the Ministry of Road and Transportation, N3 expresses the number of villages and cities around the road, N4 is the number of side-road parking lots, N5 the number of car service centers, N6 the number of medical treatment services, and N7 states the number of telephone booths (telecommunication centers) along the road segment. 2.5. The facilities/services variable For impedance modeling, the availability of several kinds of services is significant. In this study, the AHP method for the services item includes fuel stations, terminals, and public service centers and is given by Eq. (6)

T 5 ¼ ð0:134O1 þ 0:093O2 þ 0:773O3 Þ

ð6Þ

where T5 is the services variables of the road network, O1 presents the effective fuel station around road network, O2 is terminals for passengers and drivers, and O3 states the number of public service centers. The derived impedance models were implemented on an Iranian road network. In this research, the road network between Tehran and Mashhad (55,000 km) was selected as the study area. Tehran is the capital of Iran and Mashhad is one of the largest cities in the nation. In this region, many paths exist between these two cities, and a GIS road network and other features surround the road with a scale of 1:250,000. The GIS data include spatial and nonspatial data of roads and other features in the vicinity of the roads, such as facilities (e.g., fuel station), and tourist features (sea, jungle), as well as weather conditions, road traffic, and other data. For implementation of variables, the effective zone of each variable should be determined. The effective zone shows the area around each variable that can influence the calculation of the impedance model of each road segment. In this research two different phenomena could be considered regarding variables. The first includes the road traffic and the weather variables, both of which have separate effective zones for each sub-variable. The second includes the sub-models of the two main variables of security of the services. All have multiple and non-differentiable effective zones. For the second phenomenon, there are three methods of differentiating the effective zones, a buffering method, the average number of kilometers, and a special method. The effect zone of each sub-variable was determined through overlapping and intersection with each route part (Sadeghi-Niaraki et al., 2004).

12002

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

3. Model sensitivity analysis Evaluation of a multi-variable analysis includes a broad range of activities designed to assess the model’s overall reliability for various tasks. One significant technique for model evaluation is a sensitivity analysis (SA). Sadeghi-Niaraki, Kim, and Varshosaz (2010) presents other technique apart from SA for the model evaluation. A more useful model solution can be supplied by a SA (Phillips, Ravindran, & Solberg, 1976). For any multi-variable technique it is recommended that a SA be performed. A SA allows verification of the results of the decision by decision-makers to identify how changes in judgments or priority about the importance of each variable might affect outputs. (Dantzig, 1963; Evans, 1984) stated that a SA is an essential concept in effectively utilizing and implementing quantitative decision models. A SA serves different purposes in decision-making analyses (e.g., Alexander, 1989; Dantzig, 1963; Harrell, Ghosh, & Bowden, 2000; Howard, 1968; Kelton, Sadowski, & Sadowski, 1998; Reilly, 2000; Saltelli, 2004). Triantaphyllou (2000) compared various multi-variable decision making methods, including AHP. Many researches have tried to perform SAs in AHP without causing the rank reversal problem (Arbel & Vargas, 1990; Farkas, Gyorgy, & Rozsa, 2004; Moreno-Jimenez & Vargas, 1993; Sugihara & Tanaka, 2001). A SA is a particularly important aspect of an AHP problem analysis, since AHP outcomes are based on subjective expert assessments. This kind of analysis is conceived as a stage in the model evaluation that considers the extent of outcome variation of the models when variables’ weights are systematically varied over a range of interest (Qureshi, Harrison, & Wegener, 1999; Snowling & Kramer, 2001). Indeed, model inputs (variables) are subject to various sources of uncertainty including incomplete information and understanding about variables and mechanisms of a problem. This uncertainty compels a limit on the confidence in the output of the model. Thus, the SA process can help diminish uncertainty in outputs (Crosetto, Tarantola, & Saltelli, 2000). The benefits of performing a SA cover many aspects: first, it offer answers to what if questions and helps decision-makers reach consensus (Yeh, Kreng, & Lin, 2001). Second, its helps in identifying essential factors that influence the final decision (Armacost & Hosseini, 1994; Triantaphyllou & Sanchez, 1997); Third, SA is utilized to examine the stability and the sensitivity of a model with respect to changes in the priorities of the variables due to the subjectivity of expert judgments (Mészáros & Rapcsák, 1996). Fourth, it assists in visualizing the impact of alterations at policy levels on decisions at a practical level. Fifth, it generates conditions of possible rankings of decision alternatives under different circumstances (Winebrake & Creswick, 2003). Sixth, SA helps to evaluate the robustness of the suggested decision (Ho, 2004). Triantaphyllou and Sanchez (1997) set forth a considerable body of the literature on the development of the SA. However, literature on the SA for AHP is limited. From a SA theoretical point of view, some studies have been reported (Belton & Gear, 1983, 1985; Triantaphyllou, 2000) as have studies from a practical point of view (Forman & Gass, 2001; Golden, Wasil, & Harker, 1989) as well. Delgado and Sendra (2004) stated that the SA most frequently employed is based on variation of the weights of the variables implied in the process to evaluate whether it significantly alters the results gained. The two most significant elements in a sensitivity analysis are variable weights and variable values. Variable weights are more important in the SA as they are subjective numbers that are calculated from the decision maker’s judgments and have high probability for errors. If the weight alteration causes no change in the alternative rankings, errors in the estimation of variable weights can be considered trivial. If the ranking proves to be sensitive to one or more variable weights, the accuracy in

estimating weights should be investigated precisely. In practice, SA is performed by applying various weighting values for the main decision variables in this research. The main purpose of the SA here is to examine sensitivity of the routes are to changes in variable weights. This is useful in circumstances where uncertainties exist in the definition of the importance of different route related variables. In many cases, it is also important to know how the results will change if the weights are changed. As previously noted, the weighting of variables is the most important source of subjectivity in the model. Hence, a SA was conducted to evaluate how robust the model is to changes in the weights assigned to the variables in the route finding analysis. If the changed weights result in tangible qualitative and quantitative variations in the outcomes, then the model will be greatly dependent on the subjective judgment of the weights. On the contrary, if changes in the weights of the variables do not lead to major change in the results, it is concluded that the model outcomes are more objective and stable in providing an overall view of the impacts, given the model’s limitations noted above. The total weight of all the variables obtained from experts should always be equal to 1 or 100%. Thus the SA has to consider simultaneous variations in the weights of more than one variable. The SA illustrates that if a decision maker wants to change the weight of any variable by moving the associated bar up or down, the weights of other variables will also change and hence the rankings of the alternatives will change. In fact, there are many possibilities for a feasible range of variable weight deviations. The number of possible choices can be calculated from Eq. (7)

P ¼ 2n  1

ð7Þ

where P represents total possible options for changing the weights of variables and n is the number of variables. For this research, the number of all possible variations is about 4095 for five variables in the first level. For simplicity, instead of performing the aforementioned variations (4095 times), the variation ranges were grouped into two categories: local and general sensitivity ranges. The most common method, the local sensitivity range method, is to modify the weightings obtained from experts’ judgments. For the local sensitivity range method, an input variable is varied by a small amount in a range that is based on the decision maker’s judgment with regard to the priority of variables, and the corresponding change in the model output is observed. For instance, if, regarding the experts’ judgments, the priority of variable A is greater than B and variable B is superior than C, then the weight of variable B is greater than that of variable C and smaller than that of variable A. For this, from the perspective of the local sensitivity range method, the weight of B can be changed only between the weights of variable C and variable A. The variation of each variable is limited based upon the decision makers’ judgments. Moreover, the rationale behind this method is that some variables have limited weight variation and the decision maker does not have permission to alter these weights in a wide range. Because there are rules for some variables, for instance, the traffic variable, the importance of level service A must always be greater than level service B. Meanwhile, for the general sensitivity range method, a weight variable is varied across the entire range by taking the minimum priority value 0 and the maximum 1. Variable removal sensitivity is one step of the SA process and is performed by removing one or more variables at a time. The variable removal sensitivity analysis presents the significance of the variables used in the model in the assessment of these variables in the study area. In this research, SA can be structured to evaluate how sensitive the routes are to change along with the importance of the variables.

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

The literature related to SA documents the use of various types of software to examine the sensitivity of evaluation results to changes in variable weights. AHP practical implementation and performing the SA process are facilitated by programs such as expert choice software, which has a user-friendly sensitivity analysis module (Forman, 1983). In expert choice, a SA can be applied from any level in the AHP hierarchy; the software illustrates the sensitivity of alternatives to variable weight deviations by the user. In the present research expert choice software was employed, thus allowing changes to the variable weights, both graphically and interactively. However, expert choice software has limited functionality. It does not allow experts to change values at levels other than the first level of the AHP hierarchy; nor does it allow more than one change at a time. 3.1. Sensitivity analysis of the main variables In Fig. 1 the local priorities of the five main variables are represented by vertical bars above the name of the variable according to the scale on the left of the figure. The vertical bars represent the relative priorities of each variable, and they can be dragged up or down to observe the effect on the routes. The left y-axis illustrates the relative priority of each variable (as synthesized from the expert pairwise comparisons). The right y-axis represents the overall priority of each route. The OVERALL axis also shows the overall priority of each route, as read from the right axis. In other words, the variables are illustrated by vertical bars, and the routes are displayed as horizontal line graphs. The intersection of the route line graphs with the vertical variable lines represents the priority of the routes for the given variable. The Haraz route (HARAZ) appears to be superior to the other two routes for the four main variables of the model except for traffic variables, where the Semnan route (PHROZ) and Firuz-kuh route (SEM) were more positively assessed by the decision makers. The second-ranked Firuz-kuh route is superior to the Semnan route in all the main variables with the exception of facilities and traffic, where the Semnan route is more highly valued by the experts. For all main variables, the Haraz route ranks higher than the Firuz-kuh route, which in turn is superior to the Semnan route (Fig. 1). The exceptions are traffic and weather, where orders of the routes (alternatives) are changed. From the perspective of the local sensitivity range method, the models were stable with regard to changing all variable weights within the experts’ judgment boundary. These findings reveal that if there is a small degree of uncertainty in their judgment, the choices among there noted routes will not be changed.

12003

The variable removal sensitivity analysis in the assessment of the models presents the significance of the five variables in the study area. From the perspective of the general sensitivity range method, there were two considerable changes in the model in the first level. First, when the weight of the weather variable was changed from 0.195 to more than 0.380, the Firuz-kuh route moved from second priority to first priority as the best choice between Tehran and Mashahad. This means if the RMTO experts changed their judgments with regard to the importance of the weather variable or, if based on users’ preference, the weather factor has a priority exceeding 0.380, then the priority of the Firuzkuh route will be greater than that of the Haraz route. The results of the O–D surveying project reveal the correctness of this result, due to the better weather of the Firuz-kuh route in comparison with the Haraz route. Further changes occurred when the weight of traffic variable was altered from 0.214 to more than 0.599. This variation leads to a shift of the Semnan route from the third level to the first level as the best path between the two cities. The results of the O–D surveying project reveal that the Semnan route has substantially less traffic than the two other routes. Therefore, if the weight (importance) of the traffic variable is increased, then the Semnan route will take on the highest rank for selection. The results of the SA in this first level verify the suitability of the models according to the mean opinions of experts. Detail sensitivity analysis is explained on Sadeghi-Niaraki (2002). 3.2. Sensitivity analysis of the sub-variables 3.2.1. Sensitivity analysis for tourist sub-variables The Haraz route is also superior to the other two routes with regard to the tourist sub-variables, according to the results of a comparison analysis of experts’ judgment. However, Fig. 2 shows that the Firuz-kuh route is the best alternative with regard to the existence of mountainous and forest areas and the sub-variable of historical places. The Semnan route has the lowest priority for almost all the tourist sub-variables compared with the other routes. Fig. 2 depicts all tourist sub-variables at the second level before performing any changes to the weights of the variables. In the SA process, the variation of tourist sub-variables in the local sensitivity range did not cause any alteration in the final route selection results. For the general sensitivity range, by changing the weights of the variables historical place, forest areas, and mountainous areas from 0.045, 0.170, and 0.170 to 0.664, 0.670, and 0.737, respectively, the AHP software recalculates the priorities of the candidates based on this new relationship and the

Fig. 1. Sensitivity analysis for main variables.

12004

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

Fig. 2. Before performing SA for tourist sub-variables.

Fig. 3. After performing SA for tourist sub-variables.

Fig. 4. SA when ‘ski resort’ is the only sub-variable.

Firuz-kuh route moves from second priority to first (Fig. 3). One reason for this change is the high number of historical places and green and mountainous areas in the Firuz-kuh route compared to the Haraz route. In addition, the results of O–D surveying (RMTO,

2004) indicate that most drivers who are interested in these variables prefer the Firuz-kuh route over the Haraz route. The results of other sub-variables presented no changes in the priority of the paths. Furthermore, Fig. 4 illustrates that if the drivers plan to go

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

to a ski resort and this variable has high priority, then the only choice is the Haraz route. In other words, removing all sub-variables except the SKI variable leads to selection of the Haraz route as the best route and the Semnan route and Firuz-kuh route then take the same priority (Fig. 4). Overall, variation of the tourist sub-variables reveals that the priority of the Semnan route will not change from third position to first or second compared with the other routes. The O–D surveying also presents that drivers strongly do not select this path when they are interested in tourist travel. Thus, this shows the strength of the model in the third level. 3.2.2. Sensitivity analysis for facilities sub-variables With regard to the facilities sub-variables, the Haraz route is also superior to the other two routes along with the decision makers’ opinions (Fig. 5). However, the Semnan route is the best alternative concerning the existence of terminals. In the local sensitivity range, the priority of the Semnan route was modified from the third level to the second when the weight of the traffic variable was decreased from 0.773 to 0.373 (this range was inside the experts’ judgment). The reason for this result is the low traffic volume on the Semnan route. For the general sensitivity range, increasing the weight of the terminal variable from 0.093 to 0.323 as well as removing the gas station variable lead to election of the Semnan route as the best route (Fig. 6). The GIS database presents the rationale behind this change, showing the high number of terminal stations in the Semnan route as well as the effect of the gas station variable in the Haraz route and Firuz-kuh route compared with other paths. 3.2.3. Sensitivity analysis for weather sub-variables Fig. 7 shows that the Firuz-kuh route is superior to the other two routes with regard to the weather sub-variables, according to the mean opinion of experts in the AHP comparison results. There was not any deviation in the path with respect to changing

12005

the weather sub-variable weights in the local sensitivity range. For the general sensitivity range, by changing the weights of the cold weather variable and moderate weather variable from 0.378 and 0.275 to 0.260 and 0.398, respectively, the Haraz route moved from second priority to first (Fig. 8). By increasing the desert variable from 0.045 to 0.368 and 0.379, the Semnan route took higher priority than the Haraz route and the Firuz-kuh route, respectively. Similarly, by increasing the dry-cold variable from 0.128 to 0.799 and 0.820, the Semnan route moved to higher priority than the Firuz-kuh and the Haraz route, respectively. In addition, by altering the weights of the warm-humid weather variable and the drywarm weather variable, the priority of the routes did not change, since such weather is not found along these routes. Furthermore, the Semnan route is the best alternative for all weather subvariables with the exception of moderate and cold variables, where it is the worst option, as this route is placed near a large desert located far from mountainous and green areas. Given that most of the Semnan route is characterized by desert weather and relatively dry weather, the above result is reasonable. Changing and removing the other sub-variables do not have any effect on the path priorities. 3.2.4. Sensitivity analysis for traffic sub-variables The Semnan route is superior to the other two routes with regard to the traffic sub-variables because it is the best alternative with regard to service levels A and B together with the high priority of these levels of service. However, the Haraz routes and Firuz-kuh route are the best alternatives in relation to service levels D and C & E, respectively (Fig. 9). In other words, the Firuz-kuh route and Haraz routes are the best alternative when removing service levels A and B. If only service levels D is retained, the best route is the Haraz routes and if two variables (C and E) are kept, the best alternative is the Firuz-kuh route. In the SA process, the Semnan route has the lowest priority among almost all the tourist sub-variables

Fig. 5. Before performing SA for facilities sub-variables.

Fig. 6. The results after SA for facilities sub-variables.

12006

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

Fig. 7. The results of decision maker analysis for climate sub-variables.

Fig. 8. The results of SA for climate sub-variables.

Fig. 9. The results of SA for traffic sub-variables.

compared with the other routes. Changing the weights of traffic sub-variables in the local sensitivity range does not lead to any changes in the path priorities. Owing to the fact that there is a defined class for each traffic level of service, the general sensitivity range is not applicable. 3.2.5. Sensitivity analysis for security sub-variables The Haraz route is also superior to the other two routes regarding the security sub-variables, according to the mean opinions of experts. However, the Firuz-kuh route is the best alternative with

regard to the existence of telephone booths, cities and villages, road maintenance offices (Rahdar), and police offices. The Semnan route has the lowest priority in almost all the security sub-variables compared with the other routes. Regarding the different traffic sub-variables, the Semnan route displayed the weakest performance in all security sub-variables according to the mean opinions of the experts (Fig. 10). There was not any deviation in the path with respect to changing the weather sub-variable weights in the local sensitivity range. Additionally, for the general sensitivity range, by changing the weights of the police station,

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

12007

Fig. 10. Sensitivity analysis for security variable.

cities and villages, and telephone booth variables from 0.088, 0.034, and 0.017 to 0.618, 0.535, and 0.589, respectively, the Firuz-kuh route moved from second priority to first. Since most of the Firuz-kuh route is characterized by these three sub-variables, the above result can be deemed reasonable. Changing and removing the other sub-variables do not affect the path priorities. First, an analysis was conducted to account for uncertainty in the impact weights. To this end, perturbations were imposed on the original weights, and their effects on the final results were studied. This was achieved by setting two variation ranges, local and general, for each weight, generating a random number within that range, and performing a weighted linear aggregation with the new weights. The general purpose of the sensitivity analysis in this study was to identify the influence of different variable weights on the spatial pattern of the suitability index. In order to investigate the effect of the time horizon in examining the location of the most suitable areas, the sensitivity analysis was connected to the temporal factor. By means of this connection, it was possible to investigate how the most suitable area shifts, if suitability is related not only to the situation at a given moment, but increasingly connected to those characteristics, which would not change in the long run. With this in mind in the hierarchy-producing phase, the first level factors were arranged according to time-permanent and time-changeable factors. To complete all sensitivity analyses for the models, the weight of each sub-variable has been varied to highlight how changes in judgments affected the routes’ ranking. Based on the results of the sensitivity analyses for various variables and sub-variables, logical explanations for each alternative change were found. 4. Conclusion The main objective of this study was to overcome the present problems restricting the utility of most route finding analyses, which are based on mono-dimensional impedance models associated with the use of distance, time, and so on. This research presents a generic technique to weight and integrate various variables to create an impedance model for any network, and particularly for road networks. While the modeling concept described here can easily be implemented in any type of network, this work reports only on the example of impedance meddling for a road network implementation, as road networks are among the most widely used networks in GIS, as compared to similar networks such as gas or river networks. The AHP method was used in this study. The AHP based decision support system was utilized

in this study since this technique includes series of judgments, decision-making instances, and personal evaluations via a logical method. It can be said that this method is dependent on personal experiences to formulate a matter hierarchically; on the other hand, it employs logic to make final decisions. After construction of the impedance models, sensitivity analyses were performed. The sensitivity analysis provides an evaluation of the confidence in the model, possibly assessing the uncertainties associated with the modeling process and the outcomes of the model itself. It is anticipated that the use of several new techniques other than AHP that can integrate the variables found in this study would be useful. Moreover, use of these impedance models through a web GIS would greatly enhance tourism facilities and tourist infrastructure. This would contribute further to the development of the tourism industry, as a large number of users would be able to utilize the benefits of the results of the impedance models from any place in the world. Sadeghi-Niaraki and Kim (2009) expresses using ontology concept in route finding algorithm and other examples of ontology usage in GIS applications (Jung 2010, 2011). Acknowledgments This work was supported by an INHA UNIVERSITY Research Grant. The authors would also like to thank the Iranian Road Maintenance and Transport Organization. References Alexander, E. R. (1989). Sensitivity analysis in complex decision models. APA Journal (4), 323–333. Arbel, A., & Vargas, L. G. (1990). The analytic hierarchy process with interval judgments. In Proceedings of the MCDM conference. Washington, DC. Armacost, R., & Hosseini, J. (1994). Identification of determinant attribute using the analytic hierarchy process. Journal of the Academy of Marketing Science, 22(4), 383–392. Behbahani, H. (1997). Traffic engineering theory and application, traffic and transportation organization. Tehran, Iran. Belton, V., & Gear, T. (1983). On a short-coming of Saaty’s method of analytic hierarchies. Omega, 11(3), 228–230. Belton, V., & Gear, T. (1985). The legitimacy of rank reversal – A comment. Omega, 13(3), 143–144. Chou, Y. H. (1997, chap. 7). Exploring spatial analysis in geographic information systems (pp. 215–264). Chunithipaisan, S., James, P., & Parker, D. (2004). Online network analysis from heterogeneous datasets – Case study in the London train network, map asia conference 2004. Beijing, China. Crosetto, M., Tarantola, S., & Saltelli, A. (2000). Sensitivity and uncertainty analysis in spatial modelling based on GIS. Agriculture, Ecosystems and Environment, 81(1), 71–79. Dantzig, G. B. (1963). Linear Programming and Extensions. Princeton, NJ: Princeton University Press.

12008

A. Sadeghi-Niaraki et al. / Expert Systems with Applications 38 (2011) 11999–12008

Delgado, M., & Sendra, J. (2004). Sensitivity analysis in multicriteria spatial decision-making: A review. Human and Ecological Risk Assessment: HERA, 10(6), 1173–1187. Dijkstra, E. W. (1959). A note on two problems in connection with graphs. Numerische Mathematik, 1, 269–271. Dolan, A., & Aldous, J. (1993, chaps. 7, 8 and 18). Introduction to networks and algorithms. Dush, S. J., & Muhonen, G. P. (2002). The language of traffic. Nevada. Evans, J. R. (1984). Sensitivity analysis in decision theory. Decision Sciences, 15(1), 239–247. Farkas, A., Gyorgy, A., & Rozsa, P. (2004). On the spectrum of pairwise comparison matrices. Linear Algebra and its Applications, 385, 443–462. Forman, E. H. (1983). The analytic hierarchy process as a decision support system. In Proceedings of the IEEE computer society. Forman, E. H., & Gass, S. I. (2001). The analytic hierarchy process – An exposition. Operations Research, 49(4), 469–485. Golden, B. L., Wasil, E. A., & Harker, P. T. (1989). analytic hierarchy process: Applications and studies. Berlin Heidelberg, New York, London, Paris, Tokyo, Hong Kong: Springer-Verlag. Harrell, C., Ghosh, B. K., & Bowden, R. O. (2000). Simulation using ProModel. McGrawHill. HCM (1994). Highway capacity manual, special report 209. Washington, DC: Transportation Research Board. Ho, C. (2004). Strategic evaluation of emerging technologies in the semiconductor foundry industry (special case: Taiwan semiconductor foundry industry). Ph.D. dissertation. Portland, OR: Portland State University. Howard, R. A. (1968). The foundations of decision analysis. IEEE Transactions on Systems Science and Cybernetics, SSC-4(3). Husdal, J. (2000a). Network analysis – Network versus vector – A comparison study. University of Leicester, UK. Husdal, J. (2000b). An investigation into fastest path problems in dynamic transportation networks. Unpublished, coursework for the M.Sc. in GIS, University of Leicester. Jones, C. (1998). Geographical Information Systems and Computer Cartography, 225–230. chap. 13. Jun, C., Koo, J., & Koh, J. (2004). Developing a water pipe management system in Seoul using the GIS, p. 290 ff., ISPRS 2004 conference. Istanbul, Turkey. Jung, J. J. (2010). Reusing ontology mappings for query segmentation and routing in semantic peer-to-peer environment. Information Sciences, 180(17), 3248–3257. Jung, J. J. (2011). Exploiting multi-agent platform for indirect alignment between multilingual ontologies: A case study on tourism business. Expert Systems with Applications, 38(5), 5774–5780. Kelton, W. D., Sadowski, R. P., & Sadowski, D. A. (1998). Simulation with Arena (2nd ed.). McGraw-Hill. Leonard, J. D., & Oliveira, M. (2000). Towards an areawide service measure. In Proceedings of the fourth international symposium on highway capacity. Lupien, E., Moorl, H., & Dagermond, W. (1987). Network analysis in geographic information systems. Photogrametric Engineering and Remote Sensing, 53(10), 1417–1421. Mészáros, Cs., & Rapcsák, T. (1996). On sensitivity analysis for a class of decision systems. Decision Support Systems, 16, 231–240. Moreno-Jimenez, J. M., & Vargas, L. G. (1993). A probabilistic study of preference structures in the analytic hierarchy process with interval judgments. Mathematical and Computer Modeling, 17, 73–81.

Phillips, D. T., Ravindran, A., & Solberg, J. J. (1976). Operations research principles and practice. John Wiley and Sons, Inc.. Qodsipoor, H. (1999). Analytical hierarchical process. Tehran, Iran: Amikabir University. Qureshi, M. E., Harrison, S. R., & Wegener, M. K. (1999). Validation of multicriteria analysis models. Agricultural Systems, 62, 105–116. Reilly, T. (2000). Sensitivity analysis for dependant variables. Decision Sciences, 31(3), 22–51. RMTO (2004). Origin-destination surveying for transportation analyses in Iranian road networks. Iranian Road Maintenance and Transport Organization, Project report, 2, 70–90. Saaty, T. L., & Vargas, L. G. (1991). Prediction, projection and forecasting. Boston: Kluwer Academic Publishing, p. 24. Sadeghi Niaraki, A. (2002). Defining the cost model of Iran Roads Network in GIS, thesis of GIS master of science, faculty of geodesy and geomatics engineering (pp. 49–53). K.N. Toosi University of Technology. Sadeghi-Niaraki, A., & Kim, K. (2009). Ontology based personalized route planning system using a multi-criteria decision making approach. Expert Systems with Applications, 36(2), 2250–2259. Sadeghi-Niaraki, A., Kim, K., & Varshosaz, M. (2010). Multi-criteria decision-based model for road network process. International Journal of Environmental Research, 4(4), 573–582. Sadeghi-Niaraki, A. Varshosaz, M., & Behrooz. H. (2004). Implementation of the resulting cost model of road network in the geographic information system (GIS), ISPRS 2004 conference. Istanbul, Turkey. Saltelli, A. (2004). Global sensitivity analysis: An introduction. In Tutorial, international conference on sensitivity analysis (pp. 8–11). Santa Fe, New Mexico. Shadewald, J., Hallmark, K. S., & Souleyrette, R. R. (2001). Visualizing system-wide economic impacts of transportation projects. Journal of Urban Planning and Development, 127(4), 158–168. Snowling, S. D., & Kramer, J. R. (2001). Evaluating modeling uncertainty for model selection. Ecological Model, 138, 17–30. Stewart, W. (2005). Using the AHP and expert choice to support the MSC fisheries certification process. Marine Stewardship Council, Hartley McMaster Ltd. Sugihara, K., & Tanaka, H. (2001). Interval evaluations in the analytic hierarchy process by possibility analysis. Computational Intelligence, 17(3), 567–579. Thirumalaivasan, D., & Guruswamy, V. (1997). Optimal route analysis using GIS. http://www.gisdevelopment.net/application/Utility/transport/utilitytr0004a. htm. Triantaphyllou, E., & Sanchez, A. (1997). A sensitivity analysis approach for some deterministic multi-criteria decision-making methods. Decision Science, 28(1), 151–194. Triantaphyllou, E. (2000). Multi-criteria decision making methods: A comparative study. Boston, MA, USA: Kluwer Academic Publishers. Winebrake, J. J., & Creswick, B. P. (2003). The future of hydrogen fueling systems for transportation: An application of perspective-based scenario analysis using the analytic hierarchy process. Technological Forecasting and Social Change, 70, 359–384. Yamzon, C. P. (2005). California Department of Transportation GIS/level of service application. California, USA: California Department of Transportation. Yeh, J., Kreng, B., & Lin, C. (2001). A consensus approach for synthesizing the elements of comparison matrix in the analytic hierarchy process. International Journal of System Science, 32(11), 1353–1363.

ID 388460

Title Real world representation of a road network for route planning in GIS

http://fulltext.study/journal/431

http://FullText.Study

Pages 10