Rapid re-optimization via learning-enhanced column generation for vehicle routing with driver break scheduling

Ning Xue, Tianxiang Cui, Shi Cheng

Abstract: In many parts of the world, the road freight transportation sector is subject to stringent legal requirements regarding driver hours. These regulations present a significant challenge for developing practical and efficient schedules. The simultaneous optimization of vehicle routing and driver break schedules constitutes a major computational problem. In practice, many real-life vehicle routing problems require periodic planning and must adapt to sudden demand changes; this necessitates efficient re-optimization capabilities. This paper addresses this need by proposing a rapid re-optimization framework for the Vehicle Routing with Driver Break Scheduling. Our method integrates a column generation algorithm with a machine learning heuristic specifically designed for fast re-optimization. We evaluate the proposed approach under European Union Regulation (EC)561/2006 and Directive 2002/15/EC using two sets of benchmark instances, one based on a synthetic data and the other on a real-life data. The results demonstrate that the ML-enhanced approach is substantially faster than the implementation without the ML prediction component, reducing runtime by more than 96% (to just a matter of seconds), while increasing routing cost by less than 2% and yielding a final solution gap of 2.7% above the estimated lower bound.