Solving the post enrolment course timetabling problem by ant colony optimization

Analysing the effects of solution space connectivity with an effective metaheuristic for the course timetabling problem

“…Similarly, our best, mean, and worst results are all seen to outperform the results of van den Broek and Hurkins [47]. Finally, no significant difference is observed between the best and mean results of our method compared to Nothegger et al [40]; however, unlike our algorithm they have failed to achieve feasibility in a number of cases.…”

“…Table 3 gives a breakdown of the results achieved by our method using T end = 0.5 compared to the approaches of van den Broek and Hurkins [47], Cambazard et al [14], and Nothegger et al [40]. The latter two papers only list results for the first 16 instances.…”

“…Simulated annealing is then used to optimise this objective function and, surprisingly, after extensive parameter tuning W = 1 is found to provide their best results. Nothegger et al [40] have also attempted to optimise the DTF and SCC simultaneously, making use of ant colony optimisation to explore the space of valid solutions. Here the DTF and SCC are used to update the algorithm's pheromone matrices so that favourable assignments of events to rooms and timeslots will occur with higher probability in later iterations of the algorithm.…”

“…In particular, if connectivity is low then movements in the search space will be more restricted, perhaps making improvements in the objective function more difficult to achieve. From the research discussed above it is noticeable that some of the best approaches have attempted to alleviate this problem by relaxing some of the hard constraints [14,19], and/or by allowing events to be kept out of the timetable [40]. However, such methods also require mechanisms for coping with these relaxations, such as repair operators (which may ultimately require large alterations to be made to a solution), or by introducing terms into the objective function (which will require appropriate weighting coefficients to be determined, perhaps via tuning).…”

“…The first neighbourhood operator is based on those used by Lewis [30] and Nothegger et al [40]. Consider a valid solution S represented as a matrix Z |r|×|t| in which rows represent rooms and columns represent timeslots.…”