Stochastic programming models for distribution logistics, bikesharing and production management

In this thesis, we study four different problems, all characterized by the presence of uncertainty. The first two of them deal with a distribution system in which transshipment and/or backordering are allowed. For the first problem, we propose a two-stage stochastic program, we provide complexity results and we show that considering uncertainty explicitly in the model leads to better solutions with respect to the ones provided by the corresponding deterministic program, especially if limited recourse actions are admitted. For the second distribution problem, we propose a multi-stage stochastic model. As the complexity increases with the number of stages, we first derive optimal policies useful for solving two polynomially solvable cases. Then, for the general case, we show that the rolling horizon heuristic performs well by properly decomposing the time horizon. For the third problem, we derive a two-stage stochastic model to optimize the allocation and rebalancing activities in a bikesharing system. After showing the benefits of modeling uncertainty, we compare the solution of our stochastic program with the one obtained by Newsvendor model-based heuristics and with the real implemented system. For the fourth problem, we propose a two-stage stochastic programming model that quantifies the impact of worker assignment decisions to produce through an exponential stochastic learning curve. After linearizing it through a mixed integer program that can be solved efficiently, we perform a rigorously designed computational study and statistical analysis to derive tactics and managerial insights for how an organization should plan its production operations about assignment, cross-training and practicing. Finally, given the complexity of solving stochastic integer programs (even for the two-stage case), we propose a methodology in order to obtain monotonic chains of lower bounds for problems hard to be solved and we present some preliminary results based on instances from the literature.