Particle Swarm Optimization with CUDA

In recent years, particles’ optimization algorithm has highly been used as an effective method in solving complex and difficult optimization problems. Since particles algorithm is based on recurring population and it can be very inefficient in terms of the time required for implementation and speed to solve optimization problems with large-scale including ones which need a very large population to search in problem solution space. The main reason for this issue is that this algorithm optimization process requires a large number of function evaluations which are usually run serially. This article aims to implement particles’ optimization algorithm in parallel on graphics processing unit and to improve running efficiency and speed. The implementation results on the graphics processor show that the performance of this algorithm has greatly increased as to its implementation in parallel and with change in kernel implementation. In fact, in this study, implementation and velocity evaluation of particles algorithm implementation in parallel and based on CUDA framework has been investigated and compared. Then, there have been efforts to improve acceleration in this method in part and a new method will be proposed in CUDA framework to improve acceleration in particles algorithm and graphic processor setting