AI-assisted discrete-event simulation of public transport systems: a framework for rapid model development

Abstract

This study presents a systematic, modular prompt-construction framework for using Artificial Intelligence (AI) to assist simulation modelers in public transportation systems, leveraging large language models (LLMs). By using GPT-4o and Codex, we demonstrate how discrete-event simulation (DES) executable models can be rapidly generated with our five-part prompt construction framework, requiring minimal manual coding effort. The proposed framework was tested using a real-world case study of the in-campus shuttle bus transportation system at Universitas Indonesia. The empirical data from the case study served as the foundation for constructing a simulation model that can replicate a real-world situation. This data was provided to the AI via our modular prompt-construction framework, which was then used to generate the Python code for the simulation model. The generated model results were validated based on field observation data. The generated model yielded a Mean Absolute Percentage Error (MAPE) of 5.18%. These results indicate that LLM-assisted development of simulation models using our proposed framework can produce reliable simulation code that aligns with real-world public transportation behavior. The framework's applicability could provide a replicable, scalable method for rapid simulation development in public transportation planning and smart mobility contexts.