Hybrid local search techniques for the generalized balanced academic curriculum problem

The Balanced Academic Curriculum Problem (BACP) consists in assigning courses to teaching periods satisfying prerequisites and balancing students' load. BACP is included in CSPlib along with three benchmark instances. However, the BACP formulation in CSPLib is actually simpler than the real problem that, in general, universities have to solve in practice. In this paper, we propose a generalized formulation of the problem and we study a set of hybrid solution techniques based on high-level control strategies that drive a collection of basic local search components. The result of the study allows us to build a complex combination of simulated annealing, dynamic tabu search and large-neighborhood search. In addition, we present six new instances obtained from our university, which are much larger and more challenging than the CSPlib ones (the latter are always solved to optimality in less than 0.1 seconds by our techniques). For the sake of possible future comparisons, we make available through the web all the input data, our scores and results, and a solution validator.