Improvement And Application Of Particle Swarm Optimization Algorithm

With the development of economic globalization, the entrepreneurial logistics activity is changing from proprietary to outsourcing. The proposing of fourth party logistics(4PL) provides a new direction for the development of the logistics industry. 4PL is responsible for the functional integration and optimization of logistics resources, and is managers and supervisors of the logistics activities, 4PL has not only larger scale, more complex system, and more diversified problems in comparison with the third party logistics, but also put forward higher requirements on issues related to the processing algorithm. In a series of optimization problems of 4PL, path optimization is one of the most important and complicated problems.This paper first discusses the status quo of 4PL and the current problem of the path optimization. Then an improved algorithm of AEPSO-SA is proposed based on the weakness of the particle swarm optimization algorithm(PSO) in solving large-scale combinatorial optimization problems. For limiting the size or speed of the serial algorithm for solving large-scale path optimization problems, GPU parallel algorithm improve the accuracy and speed up the calculation, which based on Open CL parallel computing architecture and using of the AEPSO-SA algorithm.The specific work includes:(1) In-depth study of the PSO algorithm and improved the algorithm, AEPSO-SA algorithm is proposed. Firstly, according to the different distance of particle and the global optimal solution,particles are classified with different inertia weight being set, in order to increase the diversity of particles, and 4 classic functions are used to test the performance of the improved algorithm. Secondly, aiming at the lack of local search ability of PSO algorithm, the algorithm combines the local search strategy of simulated annealing algorithm. When better solutions after several search not being found, the algorithm search the neighborhood of the individual particles optimal solution. The performance test is verified that the AEPSO-SA algorithm have higher accuracy, compared the inertia weight linear decreasing PSO algorithm and simulated annealing algorithm.(2) Studied under the framework of Open CL, the program is parallel implemented by GPU. Analysis of the parallel feasibility of the AEPSO-SA algorithm, According to the characteristics of AEPSO-SA algorithm, design and implement the parallel AEPSO-SA algorithm based on Open CL.(3) The AEPSO-SA algorithm based on Open CL is applied to logistics distribution path optimization problem. The test proves that when the node size increases, the need to set more particles in the algorithm can guarantee the quality of solution. Set more particles can improve the accuracy; also make the algorithm time longer at the same time. Using the parallel AEPSO-SA algorithm can shorten the solving time, which has a good effect on solving the large scale optimization problems.