ABSTRACT:
Real-coded genetic algorithms (RCGAs) are a class of probabilistic optimization algorithms developed specifically for the continuous parameter optimization domain. Projection-based RCGA (PRCGA) is a hybrid RCGA that is benchmarked on a set 30 of noisy testbed functions. The recently introduced Comparison of Continuous Optimizers (COCO) methodology was used in carrying out the experiment reported in this paper. In the experiment, PRCGA was implemented as a generational RCGA with a multiple independent restart mechanism, tournament selection, α-blend crossover, nonuniform mutation and a mechanism to prevent stagnation and premature convergence. The maximum number of function evaluations (#FEs ) for each test run is 105 times the problem dimension. The results of this experiment show that PRCGA is able to solve more than half of the test functions with the dimension up to 40 with lower precision and can only solve about four test functions to the desired level of accuracy of 10-8 .
Keywords: Benchmarking, Black-box optimization, Real-coded genetic algorithm, Projection
