AI / ML
Rail Optima
Machine learning combined with constraint optimization for railway-related planning.
- Stack
- Python
- XGBoost
- CP-SAT
- Machine Learning
- Optimization
This case study is a work in progress. The overview below is accurate; architecture details, implementation notes and results will be added once they’ve been written up and verified.
Overview
Rail Optima combines two complementary techniques for railway-related planning: a gradient-boosted model (XGBoost) for the predictive part of the problem, and a CP-SAT constraint solver for the part that has to respect hard rules.
Why two techniques
Planning problems usually have two halves. Some quantities are uncertain and have to be estimated from data — that is a machine learning problem. Others are strict rules that a valid plan must never break — that is a constraint optimization problem. Rail Optima is built around using each tool for the half it is suited to, instead of forcing one approach to do both jobs.