Local search based methods for scheduling in the unrelated parallel machines environment

In many real-world situations it is necessary to make timely scheduling decisions. In most cases, metaheuristic algorithms are used to solve various scheduling problems because of their flexibility and their ability to produce satisfactory results in a short time. In recent years, several novel or hybrid metaheuristics have been proposed for scheduling problems. Although such research leads to new insights, it inevitably causes certain problems. First, it becomes unclear which methods perform best, especially if they are not properly compared with existing ones. Second, the proposed methods become increasingly complex, making them more difficult to understand and apply. The goal of this study is to investigate the possibility of defining efficient but simple iterative local search (ILS) methods for the parallel unrelated machines environment with minimisation of the total weighted tardiness. To improve the efficiency of ILS methods, several design decisions, such as the construction of the initial solution and choice of local search operators. The proposed methods have been compared with several metaheuristics, of which they achieve significantly better results. Thus, we conclude that it is not necessary to increase the complexity of metaheuristics to achieve better results. Rather, better results can be obtained with simple but well-designed local search methods. •Initial solution selection has high influence on metaheuristic performance.•Simple local search methods can easily outperform metaheuristics.•Simple metaheuristics perform better than more complex ones.•The selection of appropriate local search operators is mandatory.•More effort should be invested in local search than metaheuristic design.