Meta-heuristics for the single-machine scheduling total weighted tardiness problem

Abstract

Some general features of single-machine scheduling problems are described, and some of their structural properties are used to design local search procedures based on alternative definitions of neighbourhoods. In particular, the traditional idea of "exchanging the position of two jobs" is replaced by the idea of "exchanging jobs not apart more than a given number of positions (considered as a parameter of the algorithm)". For generating initial solutions, some traditional priority rules were tested with some degree of randomisation, introducing in general, a positive effect in the performance of the algorithms. Through a set of computational tests, the importance of the different parameters was evaluated, and their values for different meta-heuristic procedures (tabu search, and randomised local search) were tuned. Though these tests have been exhaustive only for a given problem (weighted tardiness), the results already available show these approaches are robust and flexible, and that, in general, satisfactory solutions can be obtained in an efficient way.