Smarter, Risk-Aware Route Planning for Utility Crews
Why it matters
Utility crews rarely know exactly how long each job will take. This paper blends machine learning and risk-aware optimization to plan daily routes that stay on time—even when jobs run long.
Using eight years of gas meter maintenance data, the author trains a gradient-boosted model (XGBoost) to predict how long each intervention will take, plus an estimate of uncertainty. Those forecasts feed a stochastic vehicle routing solver that:
- Builds risk buffers so routes finish on time with high probability
- Balances competing goals like completion rate and crew utilization
- Searches solutions with a multi-objective evolutionary algorithm
In tests, this data-driven approach reports about 20–25% improvements in operator utilization and job completions versus plans using fixed default times. It also validates the statistical assumptions behind its risk model.
Takeaway: don’t just predict job times—account for their uncertainty when you schedule. That’s how you get routes that are both efficient and reliable.
Paper: https://arxiv.org/abs/2601.07514v1
Paper: https://arxiv.org/abs/2601.07514v1
Register: https://www.AiFeta.com
OperationsResearch AI MachineLearning Logistics Routing Utilities XGBoost Optimization
