| Summary: | Solving the variants of the Vehicle Routing Problem (VRP) is invaluable to a huge range of businesses today. It allows for the calculation of efficient logistics routing which can drastically improve a firm’s competitiveness. As NP-hard combinatorial optimisation problems, they cannot be solved by exact methods: the solution space is simply too large for it to be evaluated in entirety in a reasonable timespan. Instead, metaheuristic algorithms, which produce reasonably good solutions much more quickly, are required.
Metaheuristics’ solution adequacy is known to be highly senstive to the problem at hand and their parameterisation notoriously difficult. This paper compares the performance of two popular metaheuristics, simulated annealing and a genetic algorithm, in solving five different VRPs. Through this it creates an idea of which is better suited to each instance, allowing practitioners to select the superior method for their needs. The results of the comparison show in four of five problems instances that one method is conclusively more effective. Walking-through and simplifying a published parameterisation framework in order to make this comparison fair also makes fine-tuning the metaheuristics more accessible to non-experts, with the pitfalls and critique of the endeavour presented.
Finally, the metaheuristics are applied to the real dataset of a taxi company. Whilst the simulations oppose findings of earlier comparisons, they provide substantial insight into how the operations of the firm being studied can be improved. Additionally, they demonstrate how other businesses may use metaheuristics to their advantage.
|
|---|