UCLA

Transportation systems are complex systems encompassing interacting components such as infrastructure, vehicles, and travelers. The collective interconnection between these components makes decision-making processes within the system extremely challenging. While modeling the interaction between these components is crucial for effective transportation planning and traffic management, conventional transportation system models often struggle with limitations in adaptability, transferability, and accurate prediction.This research starts with introducing a comprehensive agent-based modeling framework to evaluate the impact of new mobility options such as advanced vehicular technologies, evolving mobility patterns, and emerging vehicle usage behaviors, on transportation systems. The models developed within this framework provide a synthetic environment to evaluate the future impact on large-scale transportation systems from both demand and supply perspectives across various use cases, bridging existing gaps in adapting transportation systems to upcoming mobility innovations. Aiming to address the limitations of existing agent-based transportation system modeling, especially the time-consuming and resource-intensive approaches in current human mobility pattern modeling, I present a state-of-the-art Artificial Intelligence (AI)- driven human mobility pattern synthesis model framework. This model employs a novel generative deep learning approach for human mobility modeling and synthesis, using ubiquitous and openii source data. Additionally, the model can be fine-tuned with local data, enabling transferable and accurate representations of mobility patterns across different regions. The final segment of the research emphasizes the deployment of mobility AI network modeling in the real-world environment, especially the predictability for non-recurrent traffic conditions under complicated external environments. I present two deep-learning approaches for traffic state prediction in non-recurrent road conditions across large-scale networks and varying prediction time scales. The proposed traffic prediction models demonstrate superior performance, as validated by real-world data. In essence, this thesis provides significant contributions to the domain of transportation system modeling by improving adaptability, transferability, and predictability in response to evolving mobility challenges. The developed tools and algorithms pave the way for the broader, real-world implementation of intelligent agent and network modeling in transportation system research.