Defence of doctoral thesis in the field of systems and operations research, M.Sc. Juho Andelmin
This dissertation develops new formulations and algorithms for problems in green logistics, centralized resource allocation, and discrete multi-stage stochastic optimization.
Two new algorithms and a formulation are developed for the optimal routing of vehicles whose fuel consumption is monitored to prevent fuel depletion en route. This formulation is especially useful for the routing of electric vehicles. It first computes all refuel paths between each pair of customers with all possible combinations of refueling station visits, then discards those refuel paths that cannot be in the optimal solution, and finally replaces all refueling stations with the remaining refuel paths. This formulation is the first to combine routing and refueling decisions with minimal overhead. An exact algorithm that utilizes a full arsenal of sophisticated optimization techniques is developed, followed by a metaheuristic that computes near-optimal solutions fast and complements the exact algorithm. Both algorithms for serve as building blocks for future problem generalizations. Notably, the average optimality gap of the exact algorithm on 200 - 300 customer instances is about 0.67%.
The centralized resource allocation model computes efficiency scores for a collection (portfolio) of units, such as hospitals or banks, which are controlled by a central decision maker (DM). The model also helps allocate resources to these units so that the efficiency of the portfolio consisting of these units is maximized. Because the units typically have many inputs and outputs, the corresponding optimization problem has multiple objectives. Thus, all non-dominated portfolios satisfying the DM’s preferences are computed and the units’ relative resource ranges over these non-dominated portfolios are visualized. Our results indicate that relying on conventional efficiency scores in guiding resource allocation decisions can cause inefficiencies.
The Decision Programming framework is presented to solve discrete multi-stage decision-making problems arising in stochastic optimization and decision analysis with a novel formulation. It incorporates different types of endogenous uncertainties and risk measures in constraints and/or objective function. The framework offers more flexibility than standard tools in decision analysis, and it has just been adapted to solve a large-scale project portfolio selection problem with uncertain time-dependent returns.
Opponent is Professor M. Grazia Speranza, University of Brescia, Italy
Custos is Professor Ahti Salo, Aalto University School of Science, Department of Mathematics and Systems Analysis
Contact information of the doctoral student: +358403581315, [email protected]
The public defence will be organised via Zoom and on campus. Link to the event
The doctoral thesis is publicly displayed 10 days before the defence in the publication archive Aaltodoc of Aalto University.