Hybrid particle swarm optimization-constraint-based reasoning in solving university course timetabling problem

Abstract

Timetabling is a frequent problem in academic context such as schools, universities and colleges. Timetabling problems (TTPs) are about allocating a number of events (classes, examinations, courses, ect) into a limited number of time slots aiming towards satisfying a set of constraints. TTPs have also been described as a class of hard-to-solve constrained optimization problems of combinatorial nature. They are classified as constraints-satisfaction problems that intend to satisfy all constraints and optimize a number of desirable objectives. Various approaches have been reported in the literatures to solve TTP, such as graph coloring, heuristic, genetic algorithm and constraint logic programming. Most of these techniques generate feasible but not optimal solutions or results. Therefore, this research focuses on producing a feasible and yet good quality solution for university courses timetabling problem. In this thesis, we proposed a new hybrid approach by exploiting particle swarm optimization (PSO) and constraint-based reasoning (CBR). PSO is used to generate potential solutions to ensure that the algorithm is generic enough to avoiding local minima and problem dependency while utilizing a suitable fitness function. Meanwhile, CBR helps to satisfy constraints more effectively and efficiently by posting and propagating constraints during the process of variable instantiations. CBR procedures are applied to determine the validity and legality of the solution, followed by an appropriate search procedure to improve any infeasible solution and significantly reduce the search space. Results of this study have significantly proven that hybrid PSO-CBR has the ability to produce feasible and good quality solutions using real-world universities and benchmark datasets.