A genetic algorithm for service level based vehicle scheduling

In many practical applications, vehicle scheduling problems involve more complex evaluation criteria than simple distance or travel time minimization. Scheduling to minimize delays between the accumulation and delivery of correspondence represents a class of vehicle scheduling problems, where: the evaluation of candidate solutions is costly, there are no efficient schemes for evaluation of partial solutions or perturbations to existing solutions, and dimensionality is limiting even for problems with relatively few locations. Several features of genetic algorithms (GAs) suggest that they may have advantages relative to alternative heuristic solution algorithms for such problems. These include ease of implementation through efficient coding of solution alternatives, simultaneous emphasis on global as well as local search, and the use of randomization in the search process. In addition, a GA may realize advantages usually associated with interactive methods by replicating the positive attributes of existing solutions in the search process, without explicitly defining or measuring these attributes. This study investigates these potential advantages through application of a GA to a service level based vehicle scheduling problem. The procedure is demonstrated for a vehicle scheduling problem with 15 locations where the objective is to minimize the time between the accumulation of correspondence at each location and delivery to destination locations. The results suggest that genetic algorithms can be effective for finding good quality scheduling solutions with only limited search of the solution space.