The availability of refuelling locations for alternative fuel vehicles (AFVs) is an important factor that drivers consider before adopting an AFV; thus, the layout of initial filling stations for AFVs will influence the adoption of AFVs. This paper presents a training system for  optimising the layout of initial filling stations for AFVs by linking an agent-based model of the adoption of AFVs with a real city/area's  road network, as well as the city/area's social and economic background. In the agent-based model, two types of agents (driver  agents and station owner agents) interact with each other in a city/area's road network, stored in a GIS (Geographic Information  System). With simulation scenario analyses and a genetic algorithm, the training system presented in this paper can help decision  makers determine close-to-optimal layouts for initial AFV filling stations. This paper also presents  case study of the application of the  training system that analyses the layout of fast-charging or battery-changing stations for the promotion of electric vehicles adoption in Shanghai.