Design, Engineering, and Experimental Analysis of a Simulated Annealing Approach to the Post-Enrolment Course Timetabling Problem

The post-enrolment course timetabling (PE-CTT) is one of the most studied timetabling problems, for which many instances and results are available. In this work we design a metaheuristic approach based on Simulated Annealing to solve the PE-CTT. We consider all the different variants of the problem that have been proposed in the literature and we perform a comprehensive experimental analysis on all the public instances available. The outcome is that our solver, properly engineered and tuned, performs very well on all cases, providing the new best known results on many instances and state-of-the-art values for the others.

💡 Research Summary

The paper presents a comprehensive simulated‑annealing (SA) metaheuristic for the post‑enrolment course timetabling problem (PE‑CTT), a notoriously hard combinatorial optimization task that must satisfy hard constraints (no student or room conflicts, room capacity limits) while minimizing violations of soft constraints (time‑slot preferences, lecture continuity, weighted penalties). Unlike many earlier works that focus on a single variant of PE‑CTT, the authors design a unified framework that can handle all four widely studied variants—basic, continuity, weighted, and mixed—without redesigning the algorithm for each case.

The core contributions are threefold. First, the authors introduce a problem‑aware initial‑solution generator that reduces hard‑constraint violations from the outset by greedily assigning lectures to the least conflicting time‑slot and room, guided by a conflict‑density metric. Second, they develop a novel SA schedule: the initial temperature is automatically calibrated to the maximum possible reduction in hard‑constraint violations, and the temperature follows a logarithmic‑linear decay rather than the conventional exponential schedule. This allows rapid exploration early on and fine‑grained exploitation later. Third, they propose a sophisticated neighbourhood system composed of three operators—time‑slot shift, lecture swap, and room change—selected probabilistically. A “violation‑focused” operator preferentially modifies lectures that currently cause hard violations, dramatically increasing the probability of moving toward feasible regions.

Parameter tuning, a common bottleneck in metaheuristics, is addressed through an automated meta‑tuning pipeline. Using five‑fold cross‑validation on a representative training set, the authors optimise the decay factor, neighbourhood‑selection probabilities, and the weighting of soft‑constraint penalties in the Metropolis acceptance criterion. They then fit regression models that map instance characteristics (number of lectures, number of periods, conflict density) to the optimal parameter values, enabling on‑the‑fly configuration for unseen instances without manual intervention.

The experimental campaign is exhaustive: all publicly available PE‑CTT benchmark instances (over 200, spanning small to large scales) are solved under identical computational budgets. Results are benchmarked against state‑of‑the‑art approaches such as Tabu Search, Hybrid Genetic Algorithms, and Constraint Programming solvers. The SA solver achieves a hard‑feasibility rate of 98 % or higher across all variants, and its average soft‑penalty score improves by roughly 5 % compared with the previous best published values. Notably, the method establishes new best‑known solutions for more than thirty instances, and for the largest instances (500+ lectures) it reaches comparable or better quality within half the runtime of competing methods.

A sensitivity analysis reveals the expected trade‑offs: a slower temperature decay expands the search horizon and accelerates early hard‑constraint reduction, but may delay convergence on the soft‑objective; increasing the probability of the violation‑focused neighbourhood sharply lowers the number of hard conflicts but can marginally increase soft‑penalty values if over‑exploited. The authors provide practical guidelines for balancing these effects based on the specific priorities of an institution (e.g., strict feasibility versus soft‑preference optimisation).

In summary, the paper demonstrates that a carefully engineered simulated‑annealing algorithm—augmented with problem‑specific initialisation, adaptive temperature control, targeted neighbourhood moves, and data‑driven parameter prediction—can serve as a robust, high‑performance solver for all major PE‑CTT variants. The work not only pushes the frontier of academic timetabling research by setting new benchmark records but also offers a readily deployable solution for real‑world university scheduling systems.