engleski

A New Model of Global Parallel Genetic Algorithm

In this paper we describe a multithreaded parallel genetic algorithm (PGA) implementation. Considering the basic models of parallel genetic algorithms, we identify a variant of global PGA (GPGA) as the most appropriate one for use on a multiprocessor system with few processors. The difference between the synchronous and asynchronous model is analized and their characteristics are evaluated. Unlike some authors, we choose not to allow a single individual to be engaged in a tournament competition in more than one instance (no duplicates). The probability of selection for elimination of an individual is than determined based on the fitness of the chromosome and compared with the same probability of the duplicate-allowing algorithm. Finally, main advantages and disadvantages of our GPGA as well as performance comparison with sequential GA are stated.

parallel genetic algorithm; multithreading; tournament selection

nije evidentirano