Using Large Language Models to Solve the Electric Vehicle Routing Problem with Advanced Prompting Techniques

Electric vehicles are an essential technology in the energy transition that can help to reduce global carbon emissions. Electric vehicle adoption suffers from a range of issues including range anxiety. Thus, to reduce range anxiety, routes can be planned to include EV stations within the path from source and destination. The Electric vehicle routing problem is a variant of the vehicle routing problem with additional constraints. These constraints may include capacity constraints, battery charging/discharging, traffic congestion, etc. In this paper, we consider the EVRP as essentially the VRP problem with the additional constraint of passing through an EV node during a trip. We show that by carefully prompting a large language model with precise information and instructions, we can guide it to solve the EVRP problem. We experiment with different prompting techniques and our results show that explicit, instruction-driven prompts with self-checks perform the best.