A comparative study of some real-coded genetic algorithms for unconstrained global optimization

In this paper, a set of new Real-Coded Genetic Algorithms (RCGAs) with local and global exploratory search capabilities are proposed. The search capabilities are based on the inclusion of a modified crossover procedure and a new global exploratory method in RCGA. The global exploratory method is based on vector projection while the modified crossover procedure is based on a limited version of the pattern search (PS) method. These modifications are introduced to increase the efficiency and robustness of RCGAs through better local and global exploration of the search region. An experimental study of the new algorithms was carried out using a set of 57 test problems. Statistical analyses and comparisons of the new algorithms with standard real coded genetic algorithm (SRCGA) and some recent global optimization algorithms were carried out. Results obtained show that the modifications remarkably improve the performance of RCGAs across the test problems.