New crossover operators for timetabling with evolutionary algorithms.

When using an evolutionary algorithm (EA) to optimise a population of feasible course timetables, it is important that the mutation and crossover operators are designed in such a way so that they don?t produce unfeasible or illegal offspring. In this paper we present some specialised, problem specific genetic operators that enable us to do this successfully, and use these in conjunction with a steady state EA to address the University Course Timetabling Problem (UCTP). We introduce a number of different crossover operators, each of which attempts to identify useful building blocks within the timetables, and investigate whether these can be successfully propagated through the population to encourage the production of high quality solutions. We test the effectiveness of these crossover operators on twenty well-known problem instances and present the results found. Whilst the results are not state-of-the-art, we make some interesting observations on the nature of the various crossover operators and the effects that they have on the evolution of the population as a whole