Solving the Capacitated Vehicle Routing Problem with Hybrid Metaheuristic Algorithms

Abstract

It is a grave study on how to solve Capacitated Vehicle Routing Problem (CVRP) by utilizing a hybrid metaheuristic algorithm which combines clustering, genetic algorithm and local search tight. Experimental data has shown that a 6.8-7.25 percent improvement in travelling distance is on average accomplished by our hybrid algorithm over traditional heuristic solutions in benchmark problems Augerat A-n69-k9 and Golden 100. That average time can be lowered to under 10 seconds by accelerating and parallelizing the computation with the GPU, compared to over 180 seconds to solve cases of medium size (around 100 customers). A second advantage of a multi-objective approach is a better workload optimization by a reduction of the longest route length (makespan) by up to 15% that focuses on efficiency in the practice of the logistics process. At a more pragmatic level, there are boots-in-the-mud payoffs of such improvements to the logistics procedures. This 5-7 percent reduction in the distance traveled translates to an average of 25,000 in smaller fleets (100 or less vehicles) and fewer fuel expenses and overall miles flown. There are also cars that are operating at more than 90% capacity that do not allow idle mileage and idle time, which increases efficiency. It is supported by the ability to re-optimize dynamically, which makes it possible with shorter run times, making it possible to update the route on line with updated demand and traffic conditions. The resulting benefits are lower operating costs, improved service and carbon emission reduction that suits the green oriented logistics objectives. Another possible field of research is to examine the adaptive parameter tuning founded on machine learning to resolve an optimization of hybrid structure on various and large-scale CVRP problems with over 1,000 nodes. An environment in which multiple goals are aimed at and where there is inclusion of environmental goals like carbon use and normal actions provides interesting sustainability lenses. Further development of GPU driven metaheuristics is essential in extending it to dynamic and stochastic VRPs in changing and fast real time conditions. Investigation of decentralized computing paradigms could also help to solve mega-fleet VRP problems.