| Summary: | Reducing CO2 emissions in transport and logistics is currently a critical goal in the field of vehicle routing as organisations are now more willing to make strategic decisions in pursuit of sustainability. However, these strategic decisions are often taken without a clear understanding of the relevant operational costs. This thesis aims to bridge the gap between strategic decisions on CO2 emissions reduction and operational decisions on the optimal routes with respect to distance travelled. Multi-objective optimisation - which provides a framework for simultaneously pursuing the best outcomes in these disparate objectives - is the most appropriate methodology to tackle this problem.
This thesis focuses on the vehicle routing problem (VRP) with simultaneous pickup and delivery (VRPSPD). Although CO2 emissions have been considered in previous VRP research, the corresponding objectives in those cases were combined with other objectives in a single-objective function. In this study, the VRPSPD is modelled as a multi-objective optimisation problem - where the objectives are distance, and fuel consumption which is equivalent to CO2 emissions. The development of the model is influenced by a real-world problem from an industrial collaborator. A multi-objective Genetic Algorithm (GA) for VRPSPD, based on an established method (NSGA-II), is developed. A novel constructive algorithm is proposed which uses both the multi-criteria decision-making methods, TOPSIS and greedy randomisation, to generate initial solutions of better quality. The output of the NSGA-II is a Pareto front, which contains solutions that represent a trade-off between travel distance and fuel consumption which is equivalent to CO2 emission. The Pareto front visually shows the trade-off between objectives in multiple solutions. It helps the logistics manager to express their a-posteriori preferences towards the importance of particular objectives - and to choose the solution which best meets those objectives. The results from the aforementioned real-world problem demonstrate that the shortest routes do not necessarily incur the lowest fuel consumption or CO2 emissions. In the majority of problem instances, NSGA-II solutions outperform the solutions generated by the collaborator company’s commercial vehicle routing software. It is also shown that the use of the TOPSIS method to generate initial solutions does help NSGA-II to produce a better final Pareto front. In order to further assess the performance of the developed NSGA-II, its performance is compared with the state-of-the-art single-objective algorithm on a set of artificially generated problem instances. It leads to conclusions that for a smaller size problem, NSGA-II produces better result in terms of fuel consumption while results for travel distance is comparable to the benchmark instances.
The cost impact is also investigated in outsourcing decision strategy scenarios that compare asset-specific and asset-sharing outsourcing of transportation logistics from the perspective of the collaborator company. The company’s decision was based purely on strategic considerations, without any visibility of the cost impact on VRP performance. Experiments with both similar and different demand patterns and close and far customer locations concluded that asset sharing for transportation logistics could be beneficial - but any decision in this regard depends on whether the collaborator company wishes to prioritise either minimising fuel consumption or travel distance.
|
|---|