Robust university course timetabling problem subject to single and multiple disruptions

• Published in: European journal of operational research Vol. 283; no. 2; pp. 630 - 646
• Main Authors: Gülcü, Ayla, Akkan, Can
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2020
• Subjects: Bi-criteria optimization; Robustness; Simulated Annealing; Stochastic combinatorial optimization; Timetabling; Bi-criteria optimization; Timetabling; Robustness; Stochastic combinatorial optimization; Simulated Annealing
• ISSN: 0377-2217; 1872-6860
• DOI: 10.1016/j.ejor.2019.11.024

•Define the problem of robust course timetabling subject to multiple disruptions.•Developed simulated annealing algorithms for both single and multiple disruptions.•Present a novel approach for sample average approximation to handle reoptimization. University course timetables are often finalized in stages, in between which, changes in the data make the earlier version infeasible. As each version is announced to the community, it is desirable to have a robust initial timetable, i.e. one that can be repaired with limited number of changes and yielding a new solution whose quality is degraded as little as possible. We define two versions of the robust timetabling problem, first one assuming that only one lecture is disrupted (its scheduled period ceasing to be feasible) and the second one assuming multiple lectures are disrupted. The objective of the algorithms is to identify a good Pareto front defined by the solution quality (penalty associated with soft-constraint violations) and the robustness measure. Two versions of a multi-objective simulated annealing (MOSA) algorithm is developed (MOSA-SD and MOSA-SAA, for single and multiple disruptions, respectively), with the difference being in the way robustness of a solution is estimated within the MOSA algorithm. Extensive computational experiments done using the International Timetabling Competition ITC-2007 data set confirm that MOSA-SD outperforms a genetic algorithm from the literature, and MOSA-SAA outperforms MOSA-SD when there are multiple disruptions. For MOSA-SAA an innovative solution network to structure feasible solutions for a set of disruption scenarios has been developed to efficiently perform sample average approximation (SAA) calculations, which can be adopted for other stochastic combinatorial optimization problems.