EPSRC DTP Hub for Sustainable Transport: Electrical Vehicle Routing Optimization Using Machine Learning

Given the current concerns regarding the environment and global warming, reducing the use of fossil fuels and replacing them with renewable energy sources is becoming increasingly important. Government's ambition is that nearly all cars and vans on our roads are zero emission by 2035 supported by "Automated and Electric Vehicles Act, 2018". Electric vehicles are expected to play a dominant role in decarbonising the transport sector. Electric vehicles have a number of limitations which make their adoption challenging. The greatest of these is the fact that these vehicles have limited driving range meaning that they must be recharged frequently where this recharging can require a significant amount of time.

In this project we will develop novel methods for optimizing the routes taken by electrical vehicles toward minimizing detours required for recharging and the corresponding delays caused by this. Delays caused by recharging can be minimized by aligning these events as best possible with existing pauses in the transportation process. For example, if an electrical vehicle carrying goods needs to be reloaded, it may be recharged while this reloading is taking place. The methods developed in this project will be general in nature but for the purposes of this project we will focus on optimizing the transportation logistics of medium to large businesses and organizations. Transportation is usually a significant part of the cost of a product and therefore it is important that it is optimized to support the adoption of electrical vehicles.

Optimizing the above electrical vehicle routing problem is provably extremely hard making it difficult to solve exactly. Therefore, in most cases one can only hope to find a relatively good solution through the use of heuristic optimization methods. In this context, a heuristic optimization method is an optimization method which does not provably always perform well but empirically performs well in many cases. Traditionally such optimization methods are manually designed using a combination of domain knowledge and experimentation. In this work we will use machine learning methods which use large volumes of data to learn useful heuristic optimization methods. This approach is motivated by recent applications of machine learning to related optimization problems which have shown to achieve state of the art results.