Sie sind auf Seite 1von 23

A Bi-Criteria Vehicles Routing Problem

Titiporn Thammapimookkul Peerayuth Charnsethikul Industrial Engineering Department Kasetsart University, Bangkok, Thailand Abstract This paper is concerned with the Vehicle Routing Problem (VRP) arisen for an ATM scheduling, a real-world routing-scheduling problem. Generally, the main objective for the VRP is to minimize the total cost or distance of travel. From the perspective of planning and control, this objective is sometimes insufficient to provide a good practical solution. In this research we attempt to reformulate the problem with simultaneous consideration of two objectives; minimizing the total traveling time and minimizing traveling time of the longest tour. Many heuristic and optimization approach have been reviewed. A new heuristic is developed and proposed. The heuristic is an appropriate extension and modification of ClarkWright procedure. The heuristic performance was found to be efficient through a number of tests. Results obtained for the case study illustrate some improvement in both total cost and workload balancing for each tour. Introduction Nowadays, commercial banks are trying to change customer behavior from doing banking transactions through branch channel to electronic channel. Consequently, there is an increasing number of Automated Telling Machines (ATM). Two types of ATMs need to be addressed, one of which is the branch ATM, the other being the out-of-branch ATM. The branches will take care of the ATM located in their respective branches, while the out-of-branch ATMs such as those located in department store will be taken care of cash centers. Each cash center has ATMs under its responsibility. Everyday cash center need to set up a route for carrying money to fill its ATMs. The number of out of branch ATMs is very large and every day we need to randomly fill up cash. Depending on the current capacity of each ATM, many alternative decisions can be made. Now the work process decision is made by operators. Thus, the problem of ATM routing is significant. In this study, methodology Re-routing ATMs is proposed in order to maximize efficiency measured by two objectives, minimizing the total traveling time and minimizing traveling time of the longest tour simultaneously. This project will support the operator in terms of decision-making of appropriate routing, by solving a sequence of the classical Vehicle Routing Problem (VRP) to achieve both goals. VRP sometimes called Vehicle Dispatching or Delivery Problem, like its well-known cousin, the Traveling Salesman Problem (TSP), is fascinating. It is easy to describe, but difficult to solve. VRP occurs in many service systems such as delivery, customer pick-up, repair and maintenance. A fleet of vehicles, each with fixed capacity, starts at a common depot and returns to the depot after visiting locations where service is demanded. A further fascinating feature of VRP is that the basic problem can be extended into an untold number of variations, which do

occur in the real-world problem. In general, it takes a longer time to find the best routes when the number of locations is larger. Generally, the main objective for the VRP is to minimize the total cost or distance of travel. From the perspective of planning and control, this objective is sometimes insufficient to provide a good practical solution. For instance, if all vehicles have the same size and capacity, unbalanced workloads may produce unbalanced performance of the system resulting in excessive delays in the delivery of cash. The longest tour in the unbalanced workloads may produce the problem in the delivery delay. Routing the transportation problem by VRP in the objective, minimizing the total distance, can give a solution that has unbalanced workloads. In order to solve this problem, we must embrace another objective of minimizing the distance of longest tour. In terms of multi-objectives decision making, these two objectives are very difficult to optimize simultaneously because they are inconsistent in that a good solution for one objective may perform badly for the others. The major goal of this research is to develop an efficient heuristic, which solves the VRP and produces solution closed to the optimal solution of each individual objective. Problem Statement In several physical distribution problems, goods must be delivered using a fleet of vehicles from central depot to a set of geographically dispersed customers. The problem is to cluster customers into a route at minimizing the routing cost and the cost of the maximal tour and satisfies every customers demand. In addition, each vehicle must leave from and return to the central depot within the time limit. Each customer is visited exactly once by a vehicle, and the total customer demand of each route must not exceed the capacity of the vehicle operating on this route. Objectives The overall objective of the research is to develop an efficient heuristic, which solves the Bi-Criteria VRP, with applications to the ATM scheduling problem. Significance of this Study The cost of the distribution and logistics accounts for a sizable part of the total operating cost of a company. However, the cost associated with operating vehicles and crews for delivery purposes form an important component of total distribution costs. Small percentage saving in these expenses could result in a large amount of savings over a number of years. This research will provide a robust problem solving technique for a real-world, large-scale vehicle routing problem to reduce operating cost. Scope In this research, we focused on the Vehicle Routing Problems with multiple homogeneous vehicles, single depot, deterministic demands, capacity restrictions and multi-objectives. Since time windows, mixed pickups and delivery, etc. are not applicable in this problem, we do not consider these constraints in this study. Literature Reviews

1. Vehicle Routing-Scheduling Problem 1.1 Vehicle Routing Problem (VRP): The classical routing problem is defined on graph G = (V,A) where V = {v1, ,vn} is a set of vertices and A = {( v1,vj ): i j, v1, vj V} is the arc set . Vertex v1 is a depot at which is located a fleet of m identical vehicles of capacity Q. The remaining vertices represent customers. Each customer is associated with a nonnegative demand dj to be collected or delivered by a vehicle. A matrix C = (cij) is defined on A; each edge (v1, vj) is associated with a distance or travel cost cij. The VRP is to design a set of m vehicle routes of minimum total cost, each starting and ending at the depot, such that each vertex of V \ {v1} is visited exactly once by one vehicle and satisfied some side constraints. A route is a sequence of locations that a vehicle must visit along with the service it provides. The routing of vehicles is primarily a spatial problem. It is assumed that no temporal or other restrictions. Impact the routing decision except for (possibly) maximum routing length constraints. Depending on the nature of each application, VRP may possess different characteristics, which in turn decide the VRP category that problem belongs to in table 1, based on a similar enumeration by Bodin and Golden (1981), who present some broad characteristic in which various VRP may differ. The entire table may be used to provide a quick description of a routing problem. Taking different combinations of options within various characteristics on the left hand side of a table 1 result in a large number of possible problem settings. Table 1 Characteristics of Routing Problem Characteristics Possible Options 1. Size of available vehicle 2. Type of available fleet 3. Housing of vehicle 4. Nature of demand 5. Location of demand 6. Underlying network One vehicle Multiple vehicle Homogeneous (only one type of vehicle) Heterogeneous (multiple types of vehicle) Special vehicle types Single depot Multiple depots Deterministic (known) demand Stochastic demand Partial satisfaction of demand At nodes At arcs Mixed Undirected Directed Mixed Euclidean Imposed (all the same) Imposed (different route) Not imposed (unlimited capacity) Imposed (same for all routes)

7. Vehicle capacity restrictions 8. Maximum route times

9. Operation 10. Cost 11. Objective

Imposed (different for each route) Not imposed Pick ups only Drop-offs (delivery) only Mixed (pickup and delivery) Variable or routing cost Fixed operating or vehicle cost Common carrier cost Minimize total routing cost Minimize number of vehicle required Minimize utility function based on service or convenience Minimize utility function based on customer priorities

* source: Bodin and Golden (1981) 1.2 Vehicle Scheduling Problem (VSP) The Vehicle scheduling problem can be thought of as a routing problem with additional constraints related to the time various activities may be carried out. The routing problem gives special importance to the spatial characteristics of the activity. In scheduling problem, however, a time is associated with each activity. Thus, the temporal aspects of vehicle movements now have to be considered explicitly. As a result, the activities are followed in both space and time. The feasibility of an activity is also influenced by both space and time characteristics, e.g. a single vehicle could not service two locations with identical delivery or pickup time. The sequencing of vehicle activities in both space and time is at the heart of the vehicle scheduling problem (Bodin et al. 1983). Real-world constraints commonly determine the complexity of the VSP. The restrictions are: a) constraint on the length of total time or distance a vehicle may be in-service before it must return to the depot; b) the restriction that certain task can only be serviced by certain types of vehicles; c) the time allowed for vehicles to service at each location (or time windows); d) the precedence requirement of the service (such as pick-ups should be done before a delivery); e) the presence of variety of depots where vehicles may be housed. 1.3 Computation Burden: Important consideration in the formulation and solution of routing and scheduling problems is the computation burden associated with various solution techniques for these problems. The nature of the growth of computation time as a function of problem size is an issue of both theoretical and practical interest. Most routing and scheduling problems may be formulated as network problems, hence they belong to the general class of network optimization problems (Golden et al. 1981). A measure of the problem size is then available in the numbers (and/or arcs) of the resulting network. A large part of network and combinatorial problems falls into the so called NP-hard class. This class is characterized by the

property that there does not exist a polynomially bounded algorithm for any particular problem in the class. It is suggested that the effort required to solve an NP-hard problem increases exponentially with the problem size in the worst case. In the area of routing and scheduling, Lenstra and Kan (1981) provided a concise overview of complexity results known to date. As expected, most routing and scheduling problems are NP-hard. Thus, when faced with an NP-hard routing and scheduling problems, one frequently resorts to approximate heuristic procedures to obtain near optimal solutions in lieu of seeking optimal solutions. A heuristic algorithm is a procedure that uses the problem structure in a mathematical (and usually intuitive) way to provide feasible and near-optimal solutions. A heuristic is considered to be effective if the solution it provides are consistently close to the optimal solution. The effectiveness of a heuristic is judged based on the quality of its solutions and the corresponding computational time required to produce those solutions. 2. Basic Vehicle Routing Problem and Algorithms 2.1 The Traveling Salesman Problem (TSP): The traveling Salesman problem (TSP) requires the determination of a minimal cost cycle that passes through each node in the relevant graph exactly once. If costs are symmetric, that is, if the cost of travelling between two locations does not depend on the direction of travel, we have symmetric TSP, otherwise we have asymmetric or directed TSP. The TSP is well known as one of the NP-complete problems. It implies that a polynomially bounded exact algorithms are commonly used for the solution of TSPs. Several authors have proposed Branch and Bound (B&B) algorithms based on the assignment problem (AP) relaxation. In the Carpaneto and Toth (1980) B&B algorithm, the problem solved at a generic node of the search tree is a modified assignment problem in which some Xij variables are fixed at 0 or 1. Recently, a group of researchers: Bob Bixby (Rice University) and David Applegate (AT&T Bell Labs), Vasek Chvatal (Rutgers University), and William Cook (Bellcore) (1993) designed an algorithm based on branch and cut technique to determine an optimal route for the TSP of 3038 cities. The calculations for 3038 cities problem were done using parallel processing on a network of 50 workstations and required total one and on half of years of computer time. The Balas and Christofides (1981) algorithm uses a stronger relaxation than the AP relaxation: sub-tour elimination constraints are introduced into the objective function in a Langragean fashion. Using this procedure, Balas and Christopher reported optimal solutions to randomly generated 325-vertex problem in less than one minute on CDC 7600. Heuristic approaches for TSP mostly fall into three board classes tour construction procedures, tour improvement procedures, and composite procedures (Bodin et al. 1983). Tour construction procedures generate an approximately optimal tour from the distance matrix. They consist of Nearest neighbor procedure (Rosenkrantz et al. 1977), Clarke and Wright Saving (Clarke and Wright 1964), Insertion procedures (including nearest insertion, cheapest insertion, arbitrary insertion, quick insertion, greatest angle insertion), Minimal spanning tree approach (Kim 1975), and Christofides heuristic (1976).

Tour improvement procedures attempt to find a better tour given an initial tour. Tour improvement procedure mostly use branch exchange approach. The 2-opt and 3-opt heuristics were introduced by Lin (1965) and the k-opt procedure (k>=3) was presented by Lin and Kernighan (1973). Composite procedures construct a starting tour from one of the tour constructive procedure and then attempt to find a better tour using one or more of the tour improvement procedures. See, for example , Russell (1995). The multiple traveling salesman problem (M-TSP) is a generalization of the TSP that comes closer to accommodating real world problems where there is a need to account for more than one salesman (vehicle). In M-TSP, M salesman are to visit all nodes in the given network in such a way that the total distance traveled by all M salesmen is minimum. An M-TSP can always be converted into an equivalent TSP by creating M copies of the located at the same position but not connected with one another. The equivalent TSP formulations of the M-TSP were derived independently by Bellmore and Hong (1974), Orloff (1974), Svestka and Huckfeltz (1973). An overview of TSP algorithms (both exact and heuristic) can be found in Laporte et al. (1992). 2.2 The single-depot multiple-vehicle node routing problem The singledepot multiple-vehicle node routing problem (sometime called the vehicle routing problem VRP) prescribes a set of delivery routes for vehicle housed at a central depot, which services all the nodes and minimizes total distance traveled. The formulations of these problems were stated by Bodin et al. (1977). This problem can be extended by adding side-constraints such as time windows. Most solution strategies for the standard VRP can be classified as one of the following approaches: Cluster-first route-second procedures, which cluster demand nodes and / or arcs first, then design economical routes over each cluster as a second step. Example of this idea was given by Gillet and Miller (1974). Route first cluster - second procedures, which work in a reverse sequence in comparison with cluster first route second. Golden et al. (1982) provided an algorithm that typifies this approach for a heterogeneous fleet size VRP. Bienstack et al. (1993) proved that no heuristic approach can be asymptotically optimal for the generic VRP. Savings or insertion procedures, which build a solution in such a way that at each step of the procedure, a current configuration that is possibly infeasible. The alternative configuration is one that yield the largest saving in the term of some criterion function, such as total cost, or that inserts least expensively a demand entity in the current configuration into the existing route or routes. Examples of these procedures can be found in Clarke and Wright (1964) or in Solomon (1987). Improvement or exchange procedures, such as the well-known branch exchange heuristic which always maintain feasibility and strive towards optimality. Other improvement procedures were described by Potvin and Rousseau (1995), including Or-opt exchange method in which one, two, three consecutive node in a route will be removed and inserted at another location within the same or another route; k-interchange heuristic in which k link in the current routes are exchanged for k new links; and 2 opt procedure which exchanges only two edges taken from two different routes.

Mathematical programming approaches, which include algorithms that are directly based on a mathematical programming formulation of the underlying routing problem. An example of this procedure was given by Fisher and Jaikumar (1981). Christofides et al. (1981) discussed Lagrangean relaxation procedures for the routing of vehicle. Interactive optimization, which is a general purpose approach in which a high degree of human interaction is incorporated into the problem solving process. Some adaptations of this approach to VRPs are presented by Krolak et al. (1970). Exact procedures for solving VRPs, which includes specialized branch and bound (B & B), dynamic programming and cutting plane algorithms. For instance, Agarwal et al. (1989) presented a set partitioning base exact algorithms to solve VRP. Kolen et al. (1987) described a branch and bound (B & B) method for the vehicle routing problem with time windows (VRPTW). Other applications of the above algorithms for VRPs were discussed in details by Christofides et al. (1981) and Laporte (1992). Heuristic approaches , e.g. simulated annealing (SA) or tabu search (TS). For example, Brame and Simchi-Levi (1995) introduced the location-based heuristic for general routing problem, which is based on formulating the routing problem as a location problem commonly called the capacitated concentrator location problem. This location problem was subsequently solved and the solution was transforms back into a solution to the routing problem. The method incorporates many routing features into the model. Thangiah et al. (1996) presented a twophase heuristic approach to solve the VRP with backhauls and time windows constraints. First, the initial solution was obtained by using Solomons insertion heuristic procedure , and was improved through a combination of -interchanges and 2-opt exchanges to find a good solution for the problem Taillard et al. (1997) introduced a tabu search heuristic for the VRP with soft time windows. The original problem was converted into the vehicle routing problem with hard time windows by adding large penalty values, and then an exchange procedure was used to swap sequences of consecutive customers between two routes. Finally, a selecting procedure was used to find to best overall solution. 2.3 The multipledepot multiple-vehicle node routing problem The multipledepot multiple-vehicle node routing problem, or Multi-Depot VRP (MDVRP), is a generalization of the standard VRP in that the fleet of vehicles now must serve M depots rather than just one. All other constraints from the standard VRP still apply. In addition, each vehicle must leave from and return to the same depot. The realworld examples of this problem include delivery of meals, chemical products, machines industrial gases, and packaged food. Since the VRP is NP-hard, the MDVRP is also NP-hard and very difficult to solve to optimality even for relatively small size instances (Bodin et al. 1983), and only two exact algorithms were proposed for the MDVRP (Renaud et al. 1996). The first, which was presented by Laporte et al. (1984), formulated symmetric problems as integer linear programs containing degree constraints, sub-tour elimination constraints, chain barring constraints (to prevent chains of cities linking two different depots) and integrality constraints. The problems were solved by a branch and bound algorithm that initially relaxes the last three types of constraints. Optimal

solutions were reported for problems containing up to 50 customers and eight depots. The other which was proposed by Laporte et al. (1988) for asymmetric MDVRPs, first transform into an equivalent constrained assignment problem. Optimal solutions were then found by a branch and bound algorithm in which the sub-problems . Results were reported for instances containing up to 80 cities and three depots. Most existing algorithms for the MDVRP are heuristics. One of the first method is due to Tillman (1969) and used the Clarke and Wright saving criterion. The algorithm first assigns each customer to its nearest depot and constructs back and fourth routes between depots and the customers. They are gradually merged into larger routes using a saving criterion that takes into account the presence of several depots. Another heuristic based on Clarke and Wright exchange process was proposed by Sumicharast and Markham (1995). This method was considered to deal with a specific MDVRP in which the constraints of delivery of different raw materials were included. the results obtained from this approach were comparable with those obtain from this approach were comparable with those obtained using LINDO package. Gillet and Johnson (1976) proposed an assignment sweep approach which is an extension of the sweep heuristic, and solved the multiple depot problem in two stages . Firstly, customers were assigned to depots to from compact and disjoint clusters; secondly, independent single-depot VRPs were solved using the sweep heuristic. Raft (1982) presented a solution technique that can handle objectives other than route length minimization. This heuristic is a modular algorithm which decomposes the problem into smaller sub-problems. The algorithm starts with a route assignment phase. After having estimated the number of vehicles needed, the algorithm constructs of customers, each assigned to one vehicle. These clusters are not assigned to any depot and are constructed to provide a small expected length. In the next phase, each route is assigned to a depot , and then a 2-opt exchange procedure is applied to each route. Chao et al. (1993) described a multi-phase heuristic. Customers were first assigned to their closest depots. A VRP was then solved for each depot using the modified savings algorithm of Golden et al. (1977). This solution was then improved by moving customers to different depots. Renauld et al. (1996) applied a tabu search heuristic to MDVRP. Their algorithm contain two parts: (1) construction of an initial solution by assigning customers to its nearest depots and then using a heuristic to find the best route selection; and (2) using tabu search to improve the solution. 2.4 The Fleet Size and Mix Problem The question commonly faced in distribution management is what is the best composition of the fleet in order to respond to customer demand at minimal cost. In real life, the appropriate fleet need not be homogeneous and a good vehicle fleet mix is likely to yield better results. In determining the vehicle fleet composition, the fixed cost of vehicles has to be one of the major factors since it accounts for approximately 80% of the total cost associated with the vehicle (Eilon 1977).

The combined vehicle routing and vehicle fleeting composition problem can be formulated as a mixed integer linear program (Salhi and Rand 1993). Other formulations can be found in Golden et al. (1984). There are only few algorithms that have been published for this vehicle mix problem. Golden et al. (1982) were among the first to address this problem. They presented two heuristics based on the adaptation of existing routing algorithms. One is the saving technique of Clarke and Wright (1964) and the other is the giant tour algorithm developed by Gillett and Miller (1974). Ferland and Michelon (1988) formulated the vehicle mix problem and showed that the exact methods and heuristics develop for the vehicle scheduling problem with time windows and with a single type of vehicle can be extended to the multiple vehicle type problem. Three heuristic based on discrete approximation, assignment method and matching method respectively were formulated. Two exact methods used a column generation technique and a time window constraint relaxation were also proposed. An efficient heuristic for determining the composition of a vehicle fleet was proposed by Salhi and Rand (1993). The heuristic base on a perturbation procedure applied within the route in order to improve the vehicle utilization of the whole fleet. The idea here is to reallocate some customers to other routes to allow a given route to be better split or combine route to be served by a cheaper vehicle. 2.5 Multi-Criteria Vehicle Routing Problem Multi-Criteria Vehicle Routing Problem is VRP with multi-objectives, these objectives are very difficult to optimize simultaneously because they are inconsistent in that a good solution against one objective may perform badly for the others. Dean and White (1975) studied balancing workloads in machine scheduling with the approach of a modified Littles algorithm. The procedure is to continue searching until the best solution with the best balance is found (measured as minimum range). Computational results of some small problems were reported. Blair and Vasquez (1984) proposed a heuristic which is based on a node exchanging procedure which transfer a node in the maximal tour to the remaining tours in order to minimize the longest tour and the total distance. An application of this method was made to solve the VRP in a Flexible Material Handing System (FMHS). They assumed that all vehicles carried a single unit load at the time they passed through a sequence of jobs, so the VRP becomes the MTSP. Later, Blair, Charnsethikul and Vasquez (1987) developed an alternative stopping rule and a different rule for node selection in the maximal tour which is transferred to the remaining tour. The algorithm was tested at three levels of 15 nodes and 3 vehicles, 35 nodes and 4 vehicles, and 5o nodes and 5 vehicles. One hundred replications were generated at each level. Husban (1985) formulated the problem of minimizing the maximal tour of the MTSP as an integer linear program and solved it using a branch and bound technique similar to that of Dean and White (1975) for small problems. He also formulated the VRP in the FMHS as a transportation problem by an arc covering approach. Two new heuristic methods were developed. The first heuristic is based on a node covering approach, while the latter heuristic is based on an arc covering method. He reported some numerical experience using the same lower bound developed in Blair and Vasquez (1984) to measure the performance of both heurtics.

2.6 Stochastic Vehicle Routing: The Stochastic Vehicle Routing Problem (SVRP) is a generalization of the deterministic VRP. The following modification to the VRP are required: Customer demand is a random variable with a known probability distribution. Route must be design before the actual demand becomes know. The objective is to minimize expected travel distance. Since they combine the characteristics of stochastic and integer programs, SVRPs are often regarded as computationally intractable (Gendreau et al. 1996) . SVRPs are usually modeled as mixed and pure integer stochastic programs, or as Markov decision processes. All known exact algorithms belong to the first category. Exact algorithms for a number of SVRPs have been proposed by Laporte et al. (1989, 1992, 1994) and Gendreau et al. (1995). Most algorithms proposed for SVRPs are heuristic, typically adapted from methods originally designed for the deterministic case. Gendreau et al. (1996) proposed a tabu search heuristic for the SVRPs addressing both stochastic demands and stochastic customers. Other illustrations can be found in Laporte et al. (1989), or in Bertsimas (1992) and Gendreau et al. (1996) for more complicated case. 2.7 Vehicle Routing with Time Windows The Vehicle Routing Problem with time Windows (VRPTW) is VRP with the added complexity of allowable delivery times, or time window, stemming from the fact that some customers impose service deadlines and earliest service time constraints. In these problems, the spatial aspect of routing is blended with the temporal aspect of scheduling, which must be performed to ensure the satisfaction of the time window constraints. Solomon (1986) pointed out that VRPTW class is fundamentally more difficult than the VRP. Solomon (1987, 1986) designed and analyzed a variety of route construction heuristics for the VRPTW. The result of the extensive computational study reported in Solomon (1987) indicate that a sequential time-space insertion algorithm (a generalization of the Mole and Jameson (1976) approach to the VRP) has proven to be very successful in a number of important VRPTW environments. Several efficient implementations of branch exchange solution improvement procedures for VRPTW were developed in Solomon et al. (1988). Furthermore, the Or-opt procedure (Or, 1976) showed the obtainability of improved solutions of equivalent quality to those produced by 3-opt whist requiring significantly less execution time . Taillard et al. (1997) proposed a new genetic-tabu search heuristic for the VRP with soft time windows called CROSS exchange. CROSS exchange can be seen as a general form of the Or-opt and 2-opt (Potvin and Rousseau, 1995) procedure. Koskosdisidis et al. (1992) presented another methodological approach for the time constrained VRP, based on the Generalized Assignment heuristic proposed by Fisher and Jaikumar (1981). The heuristic starts with a mixed integer programming formulation which treats the time window constraints, and develops an optimization based heuristic algorithm for the resulting VRP with soft time window constraints. Ferland and Fortin (1989) introduced a heuristic approach for the vehicle scheduling problem with sliding time windows. The technique was first developed

for the problem with time windows and then extended to the case with sliding time windows. While heuristics have been found to be very effective and efficient in solving a wide range of practical size VRPTW, optimal approaches have considerably lagged behind (Solomon and Desrosier 1988). Kolen et al. (1987) extended the qpath relaxation algorithm of Christofides et al. (1981) to the problem with time windows. Jornsten et al. (1986) proposed a Langragean relaxation for the computation of a lower bound for the VRPTW.

Methodology Generally, the main objective for the VRP is to minimize the total cost or distance of travel. Sometimes the solution in above objective is good in theoretical but may not be good in practical. For example if all vehicle have the same size and capacity, the solution that has unbalanced workloads may produce unbalanced performance of the system resulting in excessive delays in delivery of material. In this research, we attempt to solve this kind of delay by formulating mathematical model with two objectives. First, we minimize the total traveling time. Second we minimize the traveling time of longest tour. The mathematical model formulation is as following. Mathematical model formulation Notation The following notation are used for the model formulation: N = number of locations to be visited M = number of tours or salesmen or vehicle Q = vehicle capacity T(L) = traveling distance for tour L T = total distance traveled in the maximal tour Cij = traveling distance from location i to location j Di = demand at location i = a vector of positive integer values which gives a set of subtour Ui elimination constraints in constraint ( )

0 otherwise The formulation The basic mathematical formulation for the Bi-Criteria VRP is an integer program with two objectives. Z1 = Min T (1)

Xijk

1 if link i,j is concluded in tour k,

N+M N+M

Z2 = Min (2) Subject to

Cij Xijk

k = 1,,M

i=1 j=1

T Cij Xijk (3)


i=1 j=1

N+M N+M

k = 1,,M

where

N+M N+M

T(L) =
N+M M

i=1 j=1

Cij Xijl
= 1

l = 1,,M j = 1,,N+M

Xijk
i=1 k=1
N+M M

(4)

Xijk
(5)

= 1

i = 1,,N+M

i=1 k=1
N+M

Xihk

i=1
N+M N+M

- Xhjk = 0 (6)
j=1

N+M

h = 1,,N+M, k = 1,,M

Di Xijk
(7)

k = 1,,M

i=1j=1

Ui - Uj + (M+N) Xijk M + N 1, i,j = 1,,N+M, i j (8)


k=1

In the above formulation Equation (1) shows the objective function of minimizing the maximal tour. Equation (2) shows the objective function of minimizing the total distance. Constraint (3) ensures that T represents the maximal tour. Constraints (4) and (5) ensure that each point is visited by one and only one vehicle. Constraint (6) represents route continuity. Constraint (7) implies that the total load of each vehicle does not exceed the vehicle capacity.

Constraint (8) represents subtour elimination constraints. As mention, the combinatorial nature and complexity of multiple objectives in the mathematical model proposed in the previous chapter. A goal program may be formulated and solved by integer programming code, but the running time is costly and there are some problems of roundoff error and propagation when the problem is large. This approach is not appropriate in the case of this application. For large vehicle routing and scheduling problem (VRP), the most efficient way to obtain the solution is using approximation method (commonly called heuristic). There are several heuristics for VRP deploying different techniques. Most heuristics, however, are the adaptation of some basic well-known procedures for each special scenario. In this research, we consider the route with saving heuristic, proposed by Clarke and Wright in 1964. Saving Algorithm Clarke and Wright (1964): suggested a simple method for optimum routing of a fleet of trucks of varying capacities used for delivery from a central depot to a large number of delivery points. Clarke and Wright have modified the original method by Dantzig and Ramser. The merchandise is homogeneous with respect to the unit of capacity. The shortest route between every two points in the system is given. It is desired to allocate loads to trucks in such manner that all the merchandise are assigned and the total mileage covered is at minimum. The procedure given is simple but effective in producing a near-optimal solution. The heuristic is called the saving algorithm which is exchange procedure in the sense that each step one node of tour is exchanged for the better set. First of all, each tour simply connects depot and 1 customer. Then it combines any 2 customers into 1 route if total demand does not exceed vehicle capacity. The saving cost due to combination is calculated and the largest is selected. Then extent the route by combining with each other customer when saving cost due to combine is calculated and the largest is selected, repeating this step until demand exceeds vehicle capacity. In the method of Clake and Wright, saving of combining 2 customers i and j into one route is calculated as: Sij = dio + doj - dij Where dij denotes travel cost from customer i to j. Customer 0 stands for the depot. An illustration of the calculated is in Figure (1).

dij
i
j

di0 d0i

d0j dj0 d0i dj0

Figure 1 Illustration of the saving calculated From the previous procedure, the algorithm can be described as follows. Step 1. If P (set of unserved customers) is empty then exit. Otherwise create a new route with starting and ending point as the depot. Select the 2 customers, largest cost saving from combination, into 1 route if total demand does not exceed vehicle capacity. Step 2. If P is empty then exit. Otherwise, select one customer from P and inserted that customer into the current route. When saving cost due to combine is calculated and the largest is selected, total demand does not exceed vehicle capacity. If no customer is selection, repeat step 1. The algorithm for Saving Heuristic is presented in the Figure (2). In this algorithm it satisfy only minimize total distance. So we produce two heuristic, modified the saving heuristic, with satisfy the objective minimize the maximal tour. 1) Arc Deleting Procedure (ADP) 2) Node Exchanging Heuristic (NEH) These heuristics attempt to reduce the longest tour and maintain minimization of the total distance. The Proposed Heuristic Methods An Arc Deleting Procedure (ADP) The ADP is an appropriate modification to Saving algorithm. For example, consider a problem with six nodes two vehicles. Suppose that the solution from the Saving algorithm is 0-1-2-3-4-0, and 0-5-6-0 where 0 represent the depot. This solution is good in term of minimizing total distance but may not be good in practical solution. To satisfy balancing workloads, a heuristic technique is applied to the problem of minimum total distance solved by the Saving algorithm. This heuristic is called the arc deleting procedure. The main goal is to reduce the maximal tour. The procedure deletes each link (i,j) where (i,j) is a sequence of node in the maximal tour, by assigning Cij (cost of traveling from node i to node j) equal to infinity, and the VRP corresponding to each deletion of link (i,j) is solved by Saving algorithm. Suppose there are k arcs contained in the maximal tour, thus producing the new k solutions. We select the best improvement to continue searching in the same way until no improvement n time.

Start

Step 1
P (set of unserved customers) << 1 No Yes End

Create a new route. Select two customer that have max saving between them.

demand in route < vehicle capacity

No

Yes Add customer i and j in to route and remove its from P

Step 2
P empty ? Yes

No

Yes

L (Set of customer which can be linked in tour) empty?

No Select the customer k from L into route that have max saving.

demand in route < vehicle capacity Yes

No

Remove customer k from L

Add customer k in to route and remove it from P

Figure 2 Saving Heuristic From the previous procedure, the algorithm can be described as follows.

Step 1. Solve the VRP by Saving algorithm. Step 2. From the solution in step 1 or 3, select the produced maximal tour and start to delete each link (i,j) in that tour and resolve the VRP by Saving algorithm corresponding to each deletion of (i,j). Step 3. If there is no improvement in that maximal tour from step 2 n times, stop and print the best solution, Otherwise, select the best solution and go to step 2. The algorithm for Arc Deleting Procedure is presented in the Figure (3). Node Exchanging Heuristic (NEH) To satisfy balancing workloads, our objective function is to minimize the longest tour instead of the total distance. We applied a heuristic technique which, is called the node exchanging heuristic. The procedure is to move each node in the maximal tour to the set of the nodes in the other tours, and the VRP corresponding each moving of node is solved by Saving algorithm twice. First solving contains node in maximal tour and second solving contains all nodes in other tours and node, which move from maximal tour. Suppose there are k nodes contained in the maximal tour, thus producing the new k solutions. We select the best improvement to continue searching in the same way until no improvement n times. Step 1. Solve the VRP by Saving algorithm. Step 2. From the solution in step 1 or 3, select the produced maximal tour and start to move each node in that tour and resolve the VRP by Saving algorithm corresponding to each move node. Step 3. If there is no improvement in that maximal tour from step 2 n times, stop and print the best solution, Otherwise, select the best solution and go to step 2. The algorithm for Node Exchanging Heuristic is presented in the Figure (4).

Start

Solve the VRP by Saving heuristic

Set the solution from Saving heuristic = Solution 1

Set Solution 3 = Solution 1


Set Solution 1 = Solution 2 and set counter =0

Set Solution 3 = Solution 2 and set counter = counter +1

Finalize set of routes

Yes

counter > n ?

No

L ( set of arc between customers of longest tour in Solution 2) empty ?

Yes

Record arc 1 in set M

No
Select one arc (arc i) from L then delete it and delete all arc in M

Solution 3 better than Solution 1 ?

No

Yes

Run Saving heuristic and remove arc i from L

No

The solution better than Solution 3 or frist running?

Yes

Set solution from running Saving heuristic = Solution 3 and set arc i = arc 1

End

Figure 3 Arc Deleting Procedure (ADP)

Start

Solve the VRP by Saving heuristic

Set the solution from running Saving heuristic = Solution 1


Set Solution 3 = Solution 1

Set Solution 1 = Solution 2 and set counter =0


Set Solution 3 = Solution 2 and set counter = counter +1
Finalize set of routes
Yes

counter > 5 ?

No

L ( set of customers of longest tour in Solution 2) empty ?

Yes

Solution 3 better than Solution 1 ?


Yes

No

No

Select one customer (customer i) from L delete it from L and N (aset of customers of longest tour in Solution 2) and then record it in M (set of customers in other tours)

Run Saving heuristic in set N and then in set M and remove customer i from M

Combine two solutions from running Saving heuristic and record customer i in set N

No

The solution better than Solution 3 or frist running?


Yes

Set solution from running Saving heuristic = Solution 3 and set arc i = arc 1

End

Figure 4 Node Exchanging Heuristic (NEH)

Test results In order to evaluate the quality of solutions obtained from the proposed heuristic, the Visual Basic program, was tested on six cash centers and three instances were randomly generated from the computer program. The details of test cases are shown in Table 2. Table 2 Detail of test cases
No. 1 2 3 4 5 6 7 8 9 Source Cash center 10 Cash center 15 Cash center 11 Cash center 5 Cash center 6 Cash center 13 Random Random Random Number of the demand node 22 28 21 22 26 21 25 75 100 Vehicle capacity No. of Vehicles 30 30 30 30 30 30 30 30 30 5 5 5 5 5 5 5 5 5

For testing of six cash centers problem, we used the travel time instance instead of the travel distance because it does contain some traffic factors. Next, the ADP,NEH and saving heuristic of Clark & Wright are coded and tested using Visual-Basic. In all cases, the running time is vary depending upon the CPU speed of computers. For the largest problem (100 demand nodes), the problem can be solved within one hour using CPU Pentium III. The result of the test can be summarized as shown in the following table 3. Table 3 The best results of all of test cases.
No. Source Number of demand node 22 28 21 22 26 21 25 75 100 Total traveling time ADP 455 433 311 305 350 171 456 1431 1895 NEH 470 433 318 267 375 172 472 1458 1904 Saving heuristic 490 452 343 346 375 177 486 1482 1946 Traveling time of longest tour ADP 195 154 150 139 180 66 192 182 178 NEH 170 135 143 113 160 69 184 178 172 Saving heuristic 245 182 223 168 205 77 242 195 196

1 2 3 4 5 6 7 8 9

Cash center 10 Cash center 15 Cash center 11 Cash center 5 Cash center 6 Cash center 13 Random Random Random

Comparing of ADP and NEH, from Table 3 if we reach in criteria of minimizing total traveling time, the solutions from ADP are almost better than the solutions from NEH. On the contrary, if we approach in criteria of minimizing traveling time of longest tour, NEH can provide the solution better than those obtained from ADP. Discussions Based upon our 9-problem test, in general, the proposed heuristic outperforms saving heuristic in terms of the total distance and longest tour. We discuss strengths and weaknesses of all algorithms in all perspectives. Generally the results from the proposed methods, the ADP and NEH, are more effectively than the existing method, saving heuristic, in terms of total traveling time and traveling time of longest tour. The reason is the ADP and NEH heuristics finds the initial solution by using the saving method then repeatedly to find the better solution. Due to the objective of minimizing the longest tour is added into the proposed methods, ADP and NEH provide a good practical solution by solving the delay problem in the delivery. Also the ADP and NEH can be adapted to handle different priority using weighting factors assignment between two criteria. In the real world, to use the results from the proposed method, we have to run all of the weight ratios shown previously. The operators should select a suitable result depending upon the situation. These situations have two case studies as following. First, the depot has only one vehicle so that the objective of the minimizing traveling time of longest tour is useless. The result, which lowest total traveling time, should be selected. Second, the depot has fleet vehicles so that the results, which have traveling time of the longest tour exceed the time limit, should not be selected. To select the remaining results, if the total cost of travel is concerned. The result, which has lowest total traveling time, should be selected. In case of the workload balancing is considered. The result, which has lowest traveling time of the longest tour, should be selected. From the result, it cannot be concluded which heuristic between ADP and NEH is the better methods. Each problem cannot enumerate which heuristic is best before we test all of cases. Consequently if we use the concept of hybrid, using the result from one method for initial solution in another method, among these methods it might give a better solution. More research following this approach should be further conducted. In general, ADP spent more computational time than NEH. But to compare with saving heuristic both methods spent much computational time. So that in practical in case of the method providing the best solution but spending much time for computation may not be a good alternative. Accordingly to choose the suitable method, computation time should be concerned. Conclusions The main purpose of this study was to develop an efficient procedure for routing problem, which can be daily used by ATM scheduling. We have reviewed many existing algorithms for the Vehicle Routing Problem. The best existing heuristic algorithms are studied and tested extensively. The best existing heuristics is saving algorithm by Clarke and Wright (1964). This heuristic it gives the

minimum total distance but in this ATM scheduling problem we can not solve the problem with this single criteria. Thus, we have to add one more the criteria, which is minimize longest tour. We have devised a new heuristics that we call Arc Deleting Procedure (ADP) and Node exchanging procedure (NEH). The ADP and NEH are an appropriate modification to Saving algorithm. These heuristics tried to reformulate the problem with simultaneous consideration of two objectives; minimizing the total traveling time and minimizing traveling time of the longest tour. The result of nine test problems shows that the ADP and NEH are more effectively than the existing method, Saving heuristic, in term of total traveling time and traveling time of longest tour. Due to the objective of minimize the longest tour is added in to propose methods, therefore ADP and NEH provide a good practical solution by solve the problem delays in the delivery. Recommendation for further study Several aspects of routing problem for the ATM scheduling that remained unsolved in this study will form interesting topics for further study. The following recommendations are made for further studies: It is observed that the travel time between any two points in the distribution system is an importance parameter, which considerably affects the service time of each job. More accurate estimation of travel time leads to more precise solution obtained from the model. In this study, the travel time between any two points in the deterministic value. For more realistic, the travel time between any two points should be in term of stochastic then the traveler time could be any value in the distribution defined. In this research we focus on homogeneous vehicle (only one type of vehicle) but in the real world, some vehicle routing problems may have multiple types of vehicle, which have different capacities. For further study, we should apply the proposed method to multiple types of vehicle. References
Balas, E. and Christofides, N. 1981. A restricted Lagrangean approach to the traveling salesman problem, Mathematical Programming, Vol.21, p.19-46. Bellmore, A. and Hong, S. 1974. Transportation of multi-salesman problem to the standard TSP, Journal ACM 21, p.500-504. Bertsimas, D.J. et.al. 1996. A vehicle routing problem with stochastic demand, Operations Research, Vol. 40, p.574-585. Bixby, B., Applegate, D., Chvatal, V. and Cook, W. 1993. World-Record Traveling Salesman Problem for 3038 Cities Solved, http://nhse.cs.rice.edu/CRPC/newsletter/jan93/news.tsp.html. Blair, L. E. and Vasquez, A. 1984. Optimal Routing of Driverless Vehicles to Support Flexible Manufacturing, Paper presented in ORSA, Dallas, TX. ________., Charnsethikul, P. and Vasquez, A. 1987. Optimal Routing of Driverless Vehicles to Support Flexible Material Handing Problem Technical Report , Department of Industrial Engineering, Texas Tech University. Bodin, L. and Golden, B. 1981. Classification in vehicle routing and scheduling, Networks, Vol.11, p.97-108. _______., Golden, B., Assad, A. and Ball, M. 1983. Routing and Scheduling of Vehicles and Crews: The State of the Art, Computers and Operations Research, Vol. 10, No. 2, p.67-211. Charnsethikul Peerayuth 1993. The Multi-Traveling Salesman Problem with

Balancing Criteria. Thai Journal of Development Administration, Vol.33, No2, p.217-229 Christofides, N., Mingozzi, A. and Toth, P. 1981. Exact algorithms for the vehicle routing problem, basses on spanning tree and shortest path relaxation, Mathematical Programming, Vol.20, p.255-282. Clarke, G., J. Wright 1964. Scheduling of Vehicles from A Central Depot to A Number of Delivery Points. Operations Research, Vol.12, p.568-581. Dean, R. H., and White, E. R. 1975. Balancing Workloads and Minimizing Set Up Costs in the Parallel Processing Shop, Operation Research Quarterly, Vol.26, p.45-53. Eilon, S. 1977. Management perspective in physical distribution, Omega, Vol. 5, p.437-462. Ferland, J.A. and Fortin, L. 1989. Vehicles scheduling with sliding time windows, European Journal of Operational Research, Vol.38, p.213-266. __________. and Michelon, P. 1989. The vehicle scheduling problem with multiple vehicle types, Journal of the Operational Research Society, Vol. 39, p.577-583. Fisher, M. and Jaikumar, R. 1981. A generalized assignment heuristic for vehicle routing, Networks, Vol.11, p.109-124. Gendreau, M., Laporte, G. And Seguin, R. 1996. A tabusearch heuristic for the vehicle routing problem with stochastic demand and customers, Operations Research, Vol.44, p.3-12. Gillett, B. and Miller, L. 1974. A heuristic algorithm for the vehicle dispatch problem, Operations Research, Vol.22, p.340-349. Gillett, B. E. and J. G. Johnson 1976. Multi-Terminal Vehicle-Dispatch Algorithm, Omega, Vol.4, No.6, p.711-718. Golden, B., Assad, A., Levy, L. and Gheysens, F.G. 1982. The fleet size and mix vehicle routing problem, Management Science & Statistics Working paper 82-020, University of Maryland at College Park. ________., Assad, A., Levy, L. and Gheysens, F.G. 1984. The fleet size and mix vehicle routing problem, Computers & Operations Research, Vol.11, p.49-66. Jornsten, K., Madsen, O. and Sorenren, B. 1986. Exact solution of the vehicle routing and scheduling problem with soft time windows by variable splitting, Working paper, IMSOR, Denmark. Husban, A. O. 1985. Balancing Routes in a Class of Vehicle Routing Problem, Ph.D. dissertation, RPI, Troy, New York. Kolen, A.W.J., A.H.G.R. Kan, H.W.J.M. Trienekens 1987. Vehicle Routing with Time Window, Operations Research, Vol.35, No.2, p.266-273. Koskosidis, Y.A., Powell, W.B. and Solomon, M.M. 1992. An optimization-based heuristic for vehicle routing and scheduling with soft time windows constraints, Transportation Science, Vol.26. p.69-85. Krolark, P., Felts, W and Marble, J. 1970. A man-machaine approach toward solving the generalized truck dispatching problem, Transportation Science, Vol.6, p.149170. Laporte, G., Y. Nobert, D. Arpin 1984. Optimal Solutions to Capacitated Multidepot VRPs, Congressus Numerantium, Vol.44, p.283-292. Laporte, G.,Y. Nobert., S. Taillefer 1988. Solving a Family of Multi-Depot Vehicle Routing and Location Routing Problems, Transportation Science, Vol.22, N0.3, p.161172. Laporte, G., Louveaux, F.V. and Mercure, H. 1992. The vehicle routing problem with stochastic travel time, Transportation Science, Vol .26, p161-170. Lenstra, J. and Kan, A.R. 1981. Complexity of vehicle routing and scheduling problems, Networks, Vol.11, p.221-227. Lin, S. 1965. Computer solutions of the traveling salesman problem, Bell System Tech. Journal 44, p.2245-2269. _____. and Kernighen, B. 1973. An effective heuristic algorithm for the traveling salesman problem, Operations Research, Vol.21, p.498-516. Mole, R.H. and Jameson, S.R. 1976. A sequential route-building algorithm applying a generalized savings criterion, Operations Research Quaterly, Vol.27, p.503-511. Orloff, C. 1974. Routing a fleet of M vehicles to/from a central facility, Networks,

Vol.4, p.147-162. Or, I. 1976. Traveling salesman type combinatorial problems and their relation to the logistics of blood Banking, Ph.D dissertation, Department of Industial Engineering and Management Science, Northwesten University, Evanston, IL. Potvin, J.K. and Rousseau, J.M. 1995. An exchange heuristic for routing problem with time windows, Journal of Operational Research Society 46: p.1433-1446. Renaud, J., G. Laporte, F. F. Boctor 1996. A Tabu Search Heuristic for the MultiDepot Vehicle Routing Problem, Computers and Operations Research, Vol.23, No.3, p.229-235. Russell, R.A. 1995. Hybrid heuristic for the vehicle routing problem with time windows, Transportation Science, Vol.29, p.156-166. Salhi, S. and Rand, G.K. 1993. Incorporating vehicle routing into the vehicle fleet composition problem, European Journal of Operations Research, Vol.66, p.313-330. Solomon, M. M. 1986. On the Worst-case Performance of Some Heuristics for the Vehicle Routing and Scheduling Problem with Time Windows Constraints. Networks, Vol.16, p.161-174. Solomon, M. M. 1987. Algorithms for the Vehicle Routing and Scheduling Problems with Time Windows Constraints, Operations Research, Vol.35, p.254-262. Solomon, M.M. and Desrosiers, J. 1988. Time windows constrained routing and scheduling problem, Transportation Science, Vol.22, p.1-13. Sumichrast, R.T, I.S. Markham 1995. A Heuristic and Lower Bound for a MultiDepot Routing Problem, Computers and Operations Research, Vol.22, No.10, p.10471056. Svestka, J. and Huckfeltz, V. 1973. Computational experience with and Msalesmen traveling salesmen algorithm, Management Science, Vol.19, p.790-799. Taillard, E., P. Badeau, M. Gendreau, F. Guertin, J.-Y. Potvin 1997. A tabu Search Heuristic for the Vehicle Routing Problem with Soft Time Windows, Transportation Science, Vol. 31, No.3, p.170-186. Thagiah, S.R., J.-Y. Potvin, T. Sun 1996. Heuristic Approaches to Vehicle Routing with Backhauls and Time Windows, Computers and Operations Research, Vol.23, No.4, p.1043-1057. Tillman, F.A. 1969. The Multiple Terminal Delivery Problem with Probabilistic Demand Transportation Science, Vol. 3, p.192-204.

Das könnte Ihnen auch gefallen