CAJ | 학술논문

CUDA技术方便程序员在GPU上进行通用计算,但并没有提供随机数产生的应用接口.为此,本文提出并实现在CUDA开发平台上并行产生均匀随机数算法,测试证明算法可行.在此基础上优化基本遗传算法,并在GPU上并行实现其所有操作,提高其运行速度和准确度;分析了种群大小和遗传代数对此算法加速比及准确度的影响,并与MATLAB工具箱进行比较.实验表明,相比MATLAB遗传算法工具箱,基于CUDA平台实现的遗传算法性能更高,准确度更好.
CUDA기술방편정서원재GPU상진행통용계산,단병몰유제공수궤수산생적응용접구.위차,본문제출병실현재CUDA개발평태상병행산생균균수궤수산법,측시증명산법가행.재차기출상우화기본유전산법,병재GPU상병행실현기소유조작,제고기운행속도화준학도;분석료충군대소화유전대수대차산법가속비급준학도적영향,병여MATLAB공구상진행비교.실험표명,상비MATLAB유전산법공구상,기우CUDA평태실현적유전산법성능경고,준학도경호.
The CUDA technology provides conveniences of general computation for programmers, but there is no application programming interface of generating random number on CUDA. Therefore, this paper presents and implements a method for parallel producing random number algorithm on CUDA, and the methods is proved feasible by testing. On this condition, we implement a parallel implementation of GA on GPU, optimize the efficiency and precision of the standard GA,analyze the influence of population size and generations of evolution to efficiency and accuracy of this algorithm. The experiment shows that compared with GA Toolbox of MATLAB,the performance and the precision of this method is better.