A Heuristic Search Algorithm for Vehicle Routing Problems and

the GIS-based Vehicle Routing System Onboard

Abstracts: Focusing on a specific vehicle routing the shortest path between two nodes, but not produce
problem that is used in onboard and embedded solutions to the problem with several different
equipments and incorporates road condition and its destinations in real time according to the change of road
change, traffic rules and several different destinations, condition. In addition, recently more and more
this paper divides the vehicle routing problem into two new-building roads and other means of transportation
parts, local routing and global routing, describes the emerge, which provide more choice for travelers and
heuristic search algorithm for routing problems onboard, ensure unblocked routes, and some traffic management
and defines evaluation functions for local routing and measures such as one-way street and the banning of turn
global routing respectively. Embedded development in crossing are widely adopted, which increase the
platform eSuperMap is applied to the construction of complexity of traveling and the hardness of the routing
vehicle routing system and key techniques of the system system development.
are researched. Furthermore, based on embedded GIS, There are several researches in areas of algorithm
the vehicle routing system is designed and realized. for VRP and onboard vehicle routing system. Many
Keywords: Vehicle routing problem, Heuristic algorithms have been proposed to find out shortest paths
Search algorithm, Embedded development, Component with several different destinations, including Willard's
GIS. tabu search(TS) algorithm, Osman's simulated annealing
algorithm[ 1], Holland's genetic Algorithms, and
1 Introduction Ombuk's hybrid algorithm which divided routes into
groups by genetic algorithm and then optimized these
With the fast development of traffic program and routes by Tabu search[2].Bent and Van Hentenryck[3]
Intelligent Traffic Systems, onboard vehicle routing minimized the number of routes and then minimized
systems have become service equipment requisite for an transportation cost by large neighborhood search.
internationalization city, which possess large potential William Ho et al. [4] focused on VRP with multiple
market share and application prospect. Onboard vehicle depots and developed a genetic algorithm (GA)
routing systems can help drivers find a reasonable hybridized with several heuristics. LIU Ai Long et al. [5]
traveling path, and provide them with optimal routing presented an automatic vehicle navigation system based
information according to their destinations. As a result, it on GPS GIS and embedded system. Pei Z. Z. et
will save drivers' time, reduce their costs, decrease traffic al.[6]discussed the optimal routing algorithm
jams and environment pollution, and bring about incorporating some typical characters of city road
enormous social and economical benefits. Vehicle networks, such as restrictions of one-way and turning.
Routing Problem (abbr. VRP) in a routing system is a Heung-Suk Hwang[7] developed an integrated model
NP-hard problem. Many mathematical models, heuristic using heuristic and genetic algorithms, and developed a
and other algorithms concerning VRP have been graphical user interface, GUI-type program, Dongeui
developed, yet some problems still exist. The vehicle GA-VRP Solver. However, the existing algorithms
routing is affected greatly by stochastic factors from cannot produce solutions to the multi-destination
vehicles, customers and the condition of routes, etc. It problems in real time according to the change of road
consists of many NP-hard problems and its models condition nor deal with some special instances, for
include many variables and constraints Now in the example one-way street and the banning of turn in
existing embedded vehicle routing systems, classical crossing[8]. Furthermore, because traditional
shortest path algorithms have been adopted, such as optimization routing algorithms are more complex and
Dijkstra algorithm, Bellman-Ford-Moore algorithm, embedded equipments with little EMS memory run
Floyd algorithm, heuristic search algorithm A* slowly, it is very difficult to find out optimal routes in
algorithm etc., while these algorithms can only present onboard terminals.
The rest of the paper is organized as follows. In
section 2, the proposed heuristic search algorithm for
Supported by the national natural science foundation of China routing problems onboard is described, evaluation
(No. 70571009 and 70171040), program of Educational functions for local routing and global routing are defined
Department of Liaoning Province (2004F020), Scientific and respectively and road characteristics are considered, such
Technology Key Research Project of Ministry of Education of
China ( No.3052), Natural Science Foundation of Liaoning as one-way street, jam and road mending. In section 3,
Province(No.2001101074), PhD Programs Foundation of the embedded vehicle routing system with several
Ministry of Education of China(No.20010141025). different destinations is designed and realized, which

- 94 -
utilizes eSuperMap components and runs on GIS (3)Weight of arc: the quantitative expression of a
platform and windows CE, and the implementation of the road or some attributes. Different road attributes can be
new system in Dalian City is discussed. Finally, chosen according to different optimal objectives, for
conclusions are given in section 4. example, road length, average traveling time of road and
average traveling cost can be treated as the weight of arc
2 A heuristic search algorithm for vehicle corresponding with the road.
routing problems onboard When nodes, arcs and weights of arcs are defined,
road network is translated into directed graph. Thus the
2.1 Problem description problem of finding the optimal route between two nodes
Optimal route problems in vehicle routing systems in road networks is converted into the shortest path
belong to shortest path problems actually. Basic problem in graph theory. Though the storage amount will
algorithms such as Dijkstra algorithm cannot be used in increase when the road network is represented by
this system directly, because the routing system works in directed graph, it is still necessary to express the road
real time and desires more efficient, algorithm for network by directed graph when the road-impedance
routing. Moreover, the digital map with huge size function related to traffic flow is adopted, since the
demands a large memory space. Since computers that traffic flows in two directions of road are different in
carry out onboard vehicle routing are limited by costs general. Furthermore, directed graph can treat some
and environment, their capabilities are restricted. unusual conditions, such as one-way street and banning
Therefore it is necessary to develop efficient routing of turn in crossing, etc.
algorithms. When the traffic flow rolls strongly in From the above description, it is clear that the
different periods of time in a day, traditional static road solution of the shortest path can be determined by the
network models and optimal routing algorithm cannot demarcating of the road weight. The corresponding road
satisfy the practical demands[9, 10]. weight can be defined by difference optimization
This paper focuses on the vehicle routing problem objectives, which is the weight of each arc if it is
with several different destinations based on embedded represented by graph. There are three kinds of
GIS onboard, which involves two typical routing optimization objectives in the routing system: the
problems: local routing and global routing. shortest travel distance, the least travel time and the least
2.1.1 Global path planning problem operational cost. The methods of defining the road
The definition of the global routing problem is to weight are as follows:
choose an optimal path that passes through every goal (I)wr=lr if the optimization objective is the
node for travelers. The routing problem can be converted shortest travel distance;
into the Traveling Salesman Problem (TSP). The global (2)wr=ta if the optimization objective is the least
path planning problem can be described as follows: travel time;
A series of nodes are represented by the set of {pl, (3)wr =ca if the optimization objective is the least
operational cost.
P2, ..., p}. The origination is po, the distance (time or Where
cost) of the optimal route between p, and pj (where, i,
j=0,1,..., n, i#&j) is d(p,, pj), which is the result of the local wr is the road weight;
routing. The problem is to find a set X={1, 2, lr is the road length;
n}(elements in set X are the number of the destination ta is the average travel time spent in the road;
ca is the average operational cost of trucks spent in
nodes) and an array N(X)={v1, v2, .. ., v,1}, where vi E X, the road;
which will minimize the global path length (total time or ta and ca are decided by several factors such as the
total cost). road level, road number, road gradient and traffic
condition etc.
MinE(N) = d(pP )+ d(pvi IPvil
i=l (1)
2.2 A heuristic search algorithm for the vehicle
Equation (1) indicates that global routing depends routing problem
on the results of computation of d(p,, pj) in local routing. Since the onboard vehicle routing runs in the
The algorithm for local routing should be simplified and embedded equipment which has limited EMS memory, it
emphasis should be placed on local routing and the is necessary to find an optimal algorithm with little EMS
treatment of traffic rules. memory when an enormous road network is treated. On
2.1.2 Local vehicle routing problem the other hand, it is very important for dynamic routing
It is very important to figure road network for system to be efficient, which requires the running of an
vehicle routing based on map. The features of road algorithm to finish in a short time. Therefore, it is
network are: necessary to reduce an algorithm's running time. To
(I)Node: the crossing of road or the end-point of achieve the above mentioned goals, the precision of the
dead ends. algorithm may be ignored. An algorithm, which uses
(2)Arc: the directed link between two nodes. little EMS memory and little time and produces an ideal
solution that has small difference with the optimal

- 95 -
solution, is suitable to the routing system. It will should be replaced by h'(j) in equation (2):
incorporate the precision of the algorithm, the storage
amount and the computing time. h (j)=min{c(j,k) V ( j, k) E A }(3)
A* algorithm, the most popular heuristic search
technique, introduces known global information when where,
selecting the next node. It estimates the distance between k: all nodes adjacent to node j;
present node and the terminal to measure the probability c(j,k): the distance from j to k;
of the node's position belonging to the optimal route, so A: the arc set of the road network.
that nodes with bigger probabilities will be selected and Table OPEN and CLOSED are used when A*
the efficiency of searching will be improved. algorithm is used to find the optimal path, where table
If the heuristic evaluation function h(v) does not OPEN preserves nodes already been created and not
compute highly the costs from current node to been checked, and table CLOSED records nodes already
destination, h(v) is called to be possibility. And then A* been visited. Both tables are realized by arrays for saving
algorithm must discover the optimal route if it exists. The EMS memory. Node fingers are reserved in the tables.
search space is smaller than that of Dijkstra algorithm. The searching algorithm for optimal path is as follows:
The search space of the A* algorithm and the improved Step 1: establish empty tables OPEN and CLOSED,
Dijkstra algorithm is shown in Fig. 1. and then add original node to table OPEN.
Step 2: the first node among the last batch of nodes
in OPEN is dequeued and saved in CLOSED.
Step 3: if the number of nodes in CLOSED equals
to that of total nodes, then the optimal route is found, go
Improved Dijkstra 1 aorith to step5.
Step 4: expand node n to create all following nodes
which are not ancestors. Calculate values of f(i) for each
oiQination destination following node, and methods for treating i are as follows:
a) if i is not in OPEN or CLOSED, then insert i to OPEN,
and array all nodes in OPEN by sort ascending, go to
step 2; b) otherwise, compare d(i) and dTotalLongth(i), if
i<dTotalLongth(i) then replace dTotalLongth(i) with d(i),
and if i is in CLOSED, go to step 2.
Step 5: trace the route from destination node back to
original node, then the optimal route will be found.
Fig.1 Search space expression The flow chart of the algorithm is shown in Fig. 2.
Evaluation function h(v) used to estimate the
possibility of a node belonging to the optimal route is the
most important parameter. Whether it is defined
reasonably will directly affect the search efficiency. In
this system, according to the practical requirement, the
evaluation functions of the global and local routing
should be defined differently.
Owing to the limitation of the total number of
destinations, the solution to the local routing problem
should be precise so that results for the entire system will
be more precise. The evaluation function in local path
searching is defined as equation (2) according to the
system's practical demand.
Wh r or not the n ber o es YES
m CLOSED equals to that of total
h(J) {imin{c(j,k)+e(k,t):V(j,k)e A}, j . t nodes?
(2)
NO
where,
k: all nodes adjacent to node j;
t: the destination node;
c(j,k): the distance from j to k;
e(k,t): the estimation of the distance from k to the
destination node;
A: the arc set of the road network.
The evaluation function should be redefined, since Fig.2 The flow chart of A* algorithm
there is no fixed destination in global path searching. h*(j)

- 96 -
3 Designs and development of the GIS-based browsing and displaying can be realized by embedding
and onboard vehicle routing system control widgets of eSuperMap into eVC. The entire
model of routing system is shown in Fig. 4.
To satisfy the demand of user's working
environment, the GIS-based and onboard vehicle routing
system is required to have the features such as
human-computer interaction, expandability and
alterability in zone etc.. The system is designed to satisfy
the requirements of consuming little information storage
space, real-time and universality, and developed in PC by
using Component GIS and ODBC. After realizing system
functions, the system is repotted to Personal Digital
Assistant (abbr. PDA) in order to extend its application
fields.
3.1 The architecture of the vehicle routing system
Developed in Windows CE, eVC++ platforms,
vehicle routing system integrates the embedded-GIS and
DB, uses the digital map as the basic database, and
provides an ideal route according to users' demands. The
system consistsof three parts: 1) Man-machine Interface: Fig.4 The entire model of the routing system
obtain users' demand and display results; 2) Digital Map
Database: obtain road data in the digital map; 3) Road 3.3 System components
condition database: obtain traffic information. The 3.3.1 Map browsing module
architecture of the vehicle routing system is shown in Fig. Map browsing module realizes these functions like
3. zooming in, zooming out and move of the map by calling
functions in eSuperMap.
Zoom in(CSeDrawParameters: :uaZoomln ): users
can control the zoom in amount and select scales.
Human-computer interface
Zoom out(CSeDrawParameters::uaZoomOut): users
User demand can control the zoom out amount and select scales.
Move(CSeDrawParameters: :uaPan): when users
move the map, it will move according to the direction
Digital map database
Map data rusults m and shift quantity.
3.3.2 Vehicle routing module
The vehicle routing module is the core of this
system. There are three basic functions in this module.
Route condition database Trafficmai >
Firstly, users choose the objective of the routing, such as
the shortest path or the least time etc.. Then the original
node and several destination nodes should be chosen in
Fig.3 The architecture of the vehicle routing system map. Lastly, the optimal path should be found in shorter
based on embedded-GIS time by real-time calculation and display in map.
3.3.3 Human-computer interface module
3.2 The entire model of the system Through the human-computer interface module
The system uses the digital map of Dalian City as users input the origination, middle nodes and
the basal database. The data of GIS which the system destinations. After the vehicle routing is finished, the
uses is divided into spatial data and attributes' data. The pathway is displayed in the interface. Human-computer
spatial data mainly contains geographic position, shape interface includes display interface and input interface in
and the links of the objects related to spatial geometric this system. The display interface is the equipment that
objects. The data of attributes describes geometric can translate electric signal into visible images in real
objects' attributes, such as their names, perimeters and time. Now the color LCD is widely used as a display
codes etc.. They are linked to the functional module of screen. Considering the practical situation of the routing
eSuperMap according to the way of interlinkage of system, the design of input interface aims at security and
objects. The road condition database, which is operational simplification. Now the touch screen is
constructed by the author, can be used to solve the widely used, which is a control facility and can be
routing problem. This database mainly contains the data connected with any displays, such as CRT, LCD or ELD.
related to the road conditions. Supported by control Users can directly choose their destinations in the digital
widgets of eSuperMap, the functions of drawing, map shown in a display with a touch screen. The system

- 97 -
will solve the vehicle routing problem according to their Tab.2 Comparison between Dijkstra algorithm and
destinations and show the results in the digital map. In heuristic search al orithm
this way it is very easy for users to manipulate the Test Typical Dijkstras Algorithm in this
system. number algorithm paper
Total Search Total Search
3.4 Disposals of the one-way street and other nodes time/s nodes time/s
unconventional conditions 1 4733 17 3637 9
2 4429 16 3379 10
Some unconventional conditions such as one-way 3 4594 17 3497 9
street and traffic jam etc. are incorporated in the system 4 4117 14 3015 8
design. The system treats roads as directed graphs. The 5 3742 12 2847 7
traffic rules are recorded in the route condition database, 6 4326 15 3374 8
0 represents impassability, 1 represents two-way streets, 7 4218 15 3139 9
2 is up-street and 3 is down-street. Traffic rules are 8 3993 14 3016 8
defined as variables (nRuleCode) and the data is read 9 4374 15 3319 10
from database when routes are searched. The traffic jam 10 4183 14 2978 9
conditions caused by traffic accident and other causes are
treated in the same way. 4 Conclusion
When the nodes that users selected cannot be found
in the digital map, the most adjacent nodes will be In this paper, the vehicle routing problem with
selected in the map database according to the several different destinations is divided into local and
longitude-latitude of the users' nodes and their defining global routing problems according to the characteristics
method is FindNearsestNode. of the embedded development and vehicle routing
systems, the improved heuristic search algorithm is
3.5 Implementation of the onboard routing system adopted to get the solution to the onboard vehicle routing
with embedded-GIS problem and the results show that it is in real time and
In the onboard vehicle routing system with precise. The problem that is difficult to get solutions to
embedded-GIS, the route condition database is created the optimal path problem with several different
for finding an ideal path based on Dalian City's digital destinations on embedded equipments is solved. The
map. Road number and road type are considered based embedded-GIS platform is applied to the system design
on the original map database. In addition, gradients of and implantation and the vehicle routing with several
routes are added to the system when solving the different destinations is realized in the embedded
least-cost-path problem since there are many gradient equipment ARM9. The system's running efficiency is
roads in Dalian City. The table structure for roads in map improved. The vehicle routing system can provide users
database is shown in Talbe . with ideal routes according to their optimization
objectives established by their preference.
Tab.1 Table structure for roads in map database Though problems occurring in the real world are
field field more complicated than those solved in the designed
field field name type length others system, the system has strong expansibility since it is
ArcID arc ID Int 4 Key developed by Object Oriented technique and convenient
ArcLength arc length Int 4 to be changed.
belong road
BelongRoadName name String 10 References
RoadType road type Int 4
road [1]Osman I H. Meta-strategy simulated annealing and
RoadDirection direction It 2 Tabu search algorithms for the vehicle routing problem.
RoadNumber road number Int 4 Annu Oper Res, 1993, 41, 77-86.
Gradient [2]Ombuki B M, Nakamura M, Osamu M.A hybrid
GradientLevel level Int 2 search based on genetic algorithms and tabu search for
vehicle routing. Brock University Technical Report:#CS-
RoadGrade Road grade Int 2 02-07, 2002,5,1-7.
[3]Bent R. and Van Hentenryck P. A two-stage hybrid
The results of the shortest path with large distance local search for the vehicle routing problem with time
produced by the proposed algorithm and the typical windows. Technical Report, CS-01-06, Brown University,
Dijkstra algorithm in Dalian City's digital map are 2001,9,1-30.
shown in Table 2. It can be concluded from table 2 that [4]William Ho , Prasanta K. Dey. A Hybrid Genetic
the proposed algorithm is more effective, which Algorithm Approach for the Multi-Depot Vehicle
decreases the number of total searched nodes by 24.58% Routing Problem. Proceedings of International
and reduces search time by 41.6%. Conference on Logistics and Supply Chain Management.
Hong Kong SAR, China, 5-7, Jan. 2006,2.

- 98 -
[5]LIU Ai long ,TIAN Yan ,CHEN Tao ,YANG Chun-lei. Proceedings of International Conference on Logistics
Research and Implementation of Sound Vehicle and Supply Chain Management. Hong Kong SAR, China,
Navigation System Based on Real time Traffic 5-7, Jan. 2006.
Restriction and Embedded System[J]. Journal of [8]Zhang Ke, Liu xiaoming, Wang xiaojing. Research on
Information Engineering University, 2005, vol6(4), route planning system for vehicle automatic navigation
99-102. [J]. System engineering, 2001,19(2),48.
[6]Pei Zhongzhe, Liu xiaoming, Ding Jian. The optimal [9]Liping Fu, Rilett L R. Expected Shortest Paths in
route algorithm in the vehicle Navigation system Dynamic and Stochastic Traffic Networks.
incorporating the limits of one-way street and banning in Transportation Research Part B, 1998:32(7).
crossing[J]. Road traffic and security, 2003(4), 26-28. [IO]Chabini I, A New Algorithm for Shortest Paths in
[7]Heung-Suk Hwang.GIS-Based VRP Solver for Supply Discrete Dynamic Networks, Proceeding of 8th IFAC
Chain Network 2-D and 3-D GIS Vehicle Routing Model. Symposium on Transportation Systems, 1997.

99