2 2.7. Experimental evaluation 55 convergence of metrics across generations for the entire population are provided in Appendix B.4. 2.7.4 Parametric analysis In our parametric analysis, we consider the population size, the multi-objective functions, the FT complexity, and the e!ect of superfluous BEs. Additionally, we assess the impact of varying the parent FT in Appendix B.5. These parameters were explored to understand their influence on computational time and convergence. We generate the failure dataset as described in Section 2.7.1. Given that the evolutionary algorithm is a stochastic process, we run our algorithmfive times for each parameter combination until convergence. Using box charts in Matlab (e.g., Figure 2.7), we represent the groups of numerical data through their quartiles. Population size Figure 2.7 presents the results of the parametric analysis when varying the population size for the m.o.f.s d and sdc, using the MPPS case study (Table 2.5). 100 200 300 400 500 Population size 0 100 200 Conv. time (min) (d) 100 200 300 400 500 Population size 0 20 40 60 (c) d sdc d sdc d sdc 100 200 300 400 500 Population size 0 0.1 0.2 (b) d sdc 100 200 300 400 500 Population size 0 0.05 0.1 (a) d s c Figure 2.7: Influence of population size (ps) on (a) εd, (b) εc, (c) εs, and (d) convergence time for the m.o.f.s sdc and d in the MPPS case study (ps =400, ng =100, uc =20). Figures 2.7(a) and 2.7(b) show that the m.o.f. sdc is more consistent with larger population sizes, with both errors (ωc andωd) generally decreasing as the population size increases. Conversely, with m.o.f. d, errors also decrease with larger population sizes, but with less consistency. Figure 2.7(c) indicates that the m.o.f. sdc consistently produces smaller FTs than m.o.f. d, often even smaller than the ground truth (i.e., ωs ⇓ 23), denoted by the horizontal red line. Figure 2.7(d) illustrates that larger population sizes exponentially increase computational time for both m.o.f.s. However, the m.o.f. sdc remains consistently faster. Multi-objective functions We evaluate all m.o.f. setups (Table 2.4) using the case studies from Table 2.5 and fixed input parameters (ps =400, ng =100, and uc =20). Figure 2.8 presents results for the case studies COVID-19, MPPS, and ddFT. Results for the case studies CSD, PT, and SMS are shown in Figure B.6 (Appendix B.6).
RkJQdWJsaXNoZXIy MjY0ODMw