GNGTS 2017 - 36° Convegno Nazionale

720 GNGTS 2017 S essione 3.3 distributed local minima. In this case the algorithms are tested on a 4-dimensional Rastrigin function, and considering (for DAGA and NGA) a population of 26 individuals divided into 2 subpopulations and a maximum number of 250 generations. The results (Fig. 1b) are represented as incremental curves showing, for each generation, the cumulative sum of tests that attained convergence. In this case we observe that, as expected, the MCA is not able to efficiently explore the model space and converges in only 7 out of 50 tests, whereas NGA and DAGA show very similar performances. In particular, this example confirms that GAs are very efficient in finding the global minimum in case of regularly distributed minima (Sajeva et al., 2017). For this example, considering a serial Matlab code running on an intel i3@1.70GHz, the average convergence time for SGA and DAGA are 0.21 s and 0.49 s, respectively. Sajeva et al. (2017) demonstrated that GAs severely suffer in case of objective functions with irregularly distributed minima. For this reason, we now compare the different algorithms on the Schwefel function: (2) Fig. 1 - a) 2-D plot of the Rastrigin function. b) Comparison of the performances of the different algorithms. Note the similar performances of NGA and DAGA. Fig. 2 - a) 2-D plot of the Schwefel function. b) Comparison of the performances of the different algorithms. Note that the NGA is strongly affected by genetic drift, whereas the DAGA algorithm converges for all the tests.