About MILES¶
MILES stands for Maryland Integrated Land-use and Everyday-travel Simulator: it follows Maryland's households from where they live and work to the miles they travel each day. It is an agent-based modelling platform that links a deep-generative synthetic population, the SILO land-use model, the MAST activity model and MATSim traffic assignment for the Baltimore region and the state of Maryland.
Where the parts come from¶
| Component | Based on | What MILES adds |
|---|---|---|
| Synthetic population | Conditional variational autoencoder (Borysov et al., 2019) | PUMA embedding and size conditioning, PUMA marginal loss, income within band, logical repair, zone placement, workplace assignment |
| Land use | SILO engine | Two-level residential location (PUMA, dwelling), car ownership, workplace and school models estimated for the region; calibration and out-of-sample test |
| Activities | ABIT code base; TASHA scheduling | MAST (Maryland Activity Scheduling and Travel model): survey distributions conditioned on the day, estimated destination, stop, outing, return-home and mode models; household car allocation; joint-tour representative |
| Traffic | MATSim | OpenStreetMap network with fixed speed and capacity rules; regional transit; income-based value of time; tolls |
Principles¶
- Estimated, not tuned. Behavioural models are maximum-likelihood estimates on microdata.
- Tested on unseen data. Survey hold-out half, out-of-sample years, independent 2024 benchmarks.
- Compared with standard methods under the same inputs and measures.
- Residents on the roads. Trucks, commercial vehicles and through traffic are outside the model.