Prediction-based optimization for dynamic transport scheduling.

This project aims at developing synergies between two important domains in computer science: Combinatorial Optimization and Machine learning, in order to efficiently solve dynamic problems in logistics and transportation. We focus on dynamic vehicle routing problems, aiming to transport goods or persons, using a set of vehicles with specific characteristics. In this type of problems, the objective is usually to find an optimal set of routes, satisfying the demand and respecting the different constraints. In a dynamic problem, all the demands are not known in advance, but requests should be handled online. Dynamic vehicle routing is usually combined with geolocalisation technologies, allowing numerous applications with a strong economical as well as social impact. We investigate how machine learning methods can aid in formulating and solving vehicle routing problems. The vehicle routing problems will be formulated as a discrete contraint optimisation problem.

Keywords:machine learning, constraint optimization, logistics, dynamic planning

Disciplines:Applied mathematics in specific fields