Application of OpenCL to Population-based Optimization Algorithms Implementation

Optimizacijski problemi čest su tip problema s kojima se čovjek susreće. Često se radi o tipu problema s prevelikim prostorom rješenja za rješavanje grubom silom. Genetski algoritam jedan je od populacijskih optimizacijskih algoritama koji takav tip problema rješava uporabom metoda sličnih prirodnoj evoluciji. U ovom radu razmatraju se različiti pristupi implementacije paralelnog genetskog algoritma u tehnologiji OpenCL, a u svrhu ubrzanja izvođenja. Ispituje se učinkovitost i nedostatci takvog pristupa. Demonstrirana je implementacija na primjeru problema pronalaska prijenosne karakteristike nekog sustava.Optimization problems are often encountered type of real-world problems. In many cases, solution space of such problems is too large to be solved by brute force. Genetic algorithm is one of population-based optimization algorithms which offers a way of tackling optimization problems based on methods and principles deduced from evolution theory. This thesis covers different implementation methods of a parallel genetic algorithm using OpenCL technology, as to improve execution performance. It considers efficiency and shortcomings of such approach and demonstrates a possible implementation on the problem of determining some system’s transfer characteristic