Vehicle allocation problem with uncertain transportation requests over a multi-period rolling horizon

This work investigates optimization techniques for a multi-period vehicle allocation problem with uncertain transportation requests revealed sequentially over a rolling horizon. Policies derived from deterministic scenarios are compared: they are generated either by simple heuristics, or by more complex approaches, such as consensus and restricted expectation algorithms, or by network flow formulations over subtrees of scenarios. Myopic and a posteriori deterministic optimization models are used to compute bounds allowing for performance evaluation and for estimating the value of information. The economic benefit of the stochastic model is highlighted: our results show that the the information about future, uncertain orders contained in the stochastic part of the horizon can be used to generate improved profits. Robustness against misspecified probability distributions is examined. Subtree formulations produce the best results, are robust and can be solved efficiently, which makes them appropriate for industrial implementations.