Improving Local Search for Fuzzy Scheduling Problems

The tests have been performed on 50 problem instances generated based on the job characteristics of the practical case in the Sherwood Press, and the results have been averaged. Results Conducting significance tests at a level of 0.99, it has been possible to obser ve that for the single criterion formulation EX leads in 90% of the instances to better results than BSH, which is found to be better than FSH. The results are stable over time, thus the altering of the weights does not lead to significantly different results. The comparison of the neighbourhood search on the bicriteria extension of the problem with the results of the single criterion version reveals an interesting pattern, shown in Figure 1. After approximately 1,000 initial iterations, the single criterion search is able to significantly outperform the bicriteria formulation in more problem instances than vice versa. However, this effect is reversed after the local search approaches have been allowed more evaluations. While the monocriterion hillclimber tends to get stuck in local optima, its bicriterion counterpart successfully overcomes local optimality, leading to significantly better results. As Figure 1 reveals, this effect is each time repeated after changing the weight settings. While the neighbourhood search for the single criterion problem allows comparably fast improvements, its advantage over the multi criterion extension is decreasing over time, and the results are reversed given the necessary amount of iterations.