Parallel Adaptive Genetic Algorithm

In this paper we introduce an efficient implementation of asynchronously parallel genetic algorithm with adaptive genetic operators. The classic genetic algorithm paradigm is extended with parallelization on one hand and an adaptive operators method on the other. The parallelization of the algorithm is achieved through multithreading mechanism, a very effective and easy to implement technique. With parallelization we can get a better program structure and a significant decrease in computational time on a multiprocessor system. The adaptive method presented here determines the way in which the genetic operators are applied, not interfering with the operators themselves. It uses certain population characteristic values to estimate the diversity of the solutions in problem space and acts accordingly either to prevent premature convergence or to exploit the promising areas. The improvement we achieve with adaptation is twofold: the designed algorithm performs better over a range of domains and the user is also relieved of the task of defining its parameters. The described parallel adaptive genetic algorithm (PAGA) is applied to optimization of several multimodal functions with various degrees of complexity, employed earlier for comparative studies. Furthermore, a non-uniform mutation operator is introduced in this work and its influence on algorithm's performance is recognized.