This paper considers the deterministic vehicle routing problem with service time requirements for delivery. Service requests are also available at different times known at the initial time of route planning. This paper presents an approach based on the generation of service sequences (routes) using randomness. Both the single-vehicle and the multiple-vehicle cases are studied. Our approach is validated using random-generated data and compared against the optimal solution obtained by mathematical programming for small-sized instances, as well as against known lower bounds for medium to large-sized instances. Results show that our approach is competitive with reference to the value of the objective function, and requires less computational time in comparison with the exact resolution procedure.
INTRODUCTIONThe vehicle routing problem (VRP) has received an immense attention from the scientific community during the last three to four decades. Since the first time the problem was presented in the scientific community by Dantzig and Ramser (1959), it plays a vital role in the design of distribution systems. Basically, the classical VRP consists of designing routes for a set of vehicles that are to requested to service at the lowest cost a set of geographically dispersed customers. In order to approach the basic version of the problem to more real-life contexts, several restrictions have been added: service time windows, pickup and deliveries, backhauls, etc. (Cordeau et al. 2005). The most widely studied vehicle routing problems are the Capacitated Vehicle Routing Problem (CVRP) and the Vehicle Routing Problem with Time Windows (VRPTW). These are surveyed by Laporte and Semet (2002), and Cordeau et al. (2002a). This paper considers the classical Vehicle Routing Problem (VRP) with multiple uncapacitated homogenous vehicles (Laporte 1992, Toth and Vigo 2001, Golden, Raghavan and Wasil 2008). Formally, the classical VRP is defined on an undirected graph = ( , ) where = { 0 , 1 , … , } is a vertex set and = {� , �: , ∈ , < } is an edge set. Vertex 0 is a depot at which are basedidentical (homogeneous) vehicles of capacity , while the remaining vertices represent customer or clients. A non-negative cost, distance or travel time matrix = ( ) is defined on . Each customer has a nonnegative demand and a non-negative service time . The VRP consists on designing a set of vehicle routes of least cost, each starting and ending at the depot, and such that each customer is visited exactly once by a vehicle, and the total demand of any route does not exceed the load capacity of vehicles. The VRP is a hard combinatorial optimization problem. Exact solution procedures have been proposed (Naddef and Rinaldi 2002, Baldacci, Hadjiconstantinou and Mingozzi 2004) but they can only solve relatively small instances and their computational time are highly variable. To this day, heuristics remain the only reliable approach for the solution of practical instances. Several literature surveys have been recently published on the use of heuristics a...