The Research And Application Of Parallel Particle Swarm Optimization Algorithm Based On CUDA

PSO (Particle Swarm Optimization) is a typical swarm intelligence optimization algorithm, commonly used in solving function optimization, combinatorial optimization, neural network training and other issues. Compared with traditional optimization algorithms, PSO has a simple structure, less parameters, easy to implement and strong optimization capability and so on. However, with the development and application of modern computer technology, the problem with a large number of dimensions or high complexity often been proposed, PSO usually requires a lot of time to optimize them, and sometimes can not get right results. Therefore, finding a highly efficient parallel implementation is very important.Currently, parallel PSO algorithm is mainly based on CPU platform, the thread concurrency restrict by the number of CPU processing cores, in dealing with the complex problems caused by large-scale speed is limited. In recent years, under the promotion of video games, computer simulation and other industries, Graphics Processor Unit (GPU) computing performance far exceeds the CPU. How to use the GPU as a general purpose computing device, to accelerate the current program execution became a research hotspot. CUDA (Compute Unified Device Architecture) is NVIDIA launched a GPU-based general-purpose parallel computing solutions. It will massively parallel computational logic expressed by means of a simple, effective and affordable way to become a key leverage GPU-accelerated parallel computing implementation procedures.This paper based on CUDA parallel technologies, focusing on how to get the parallel design and implementation of PSO algorithm. The main findings include:(1) Studied the PSO algorithm for parallel granularity of analysis. Redesigned data storage structure of particle for CUDA multithreaded environment. Optimized the PSO algorithm processes of asynchronous parallel. And solved the multi-threaded communication problems. The experimental results show that sloving these complex problems which requires a lot of iterations, the acceleration has a very significant effect if based on the CUDA parallel PSO algorithm.(2)Studied the Otsu algorithm of image segmentation. Aim at the2d-Otsu algorithm cost too much time to find the optimal threshold. In this paper, combining with2d-Otsu and CUDA-PSO which was proposed in chapter3of this paper, and saving the time of finding optimal threshold. From the perspective of a typical image test results,2d-Otsu algorithm based on CUDA-PSO is guaranteed at the same time a good segmentation effect, and improve the speed of image segmentation.