Otimização de grades horárias acadêmicas com GRASP e Simulated Annealing: um estudo de caso em um Instituto Federal

Abstract

This paper presents a hybrid solver for the academic timetabling problem in a Federal Institute. The proposed approach combines a GRASP-inspired constructive heuristic with Simulated Annealing, neighborhood operators, and automatic hyperparameter calibration. The problem is modeled through a penalty-based objective function that accounts for hard and soft constraints, including teacher and class conflicts, unavailability, pedagogical limits, and timetable quality criteria. Three real-data-based instances with 127, 255, and 382 weekly class blocks were evaluated through 30 computational repetitions using distinct pseudorandom seeds. The results indicate that the solver generated feasible timetables without hard-constraint violations, with low variation in final costs and execution times compatible with institutional use. The commercial software adopted as an operational reference did not produce, under the tested configurations, complete timetables without substantial manual intervention, highlighting the relevance of customized optimization tools.