改进粒子群神经网络在出入境货物系统中的研究

发布时间：2018-01-27 04:26

本文关键词： 粒子群精英相交基学习重构初始化神经网络出入境货物风险评价　出处：《北京林业大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着经济贸易的快速发展,贸易量的大幅增加给出入境货物抽检带来挑战,原先采取的人工均匀随机布控抽检耗费人力物力,且不能有效甄别出不合格产品。考虑到布控抽检所要求的更高速度、更高精准度,给出出入境货物风险评价系统以实现智能化抽检。作者提出基于相交基学习的重构型精英粒子群算法训练神经网络以优化出入境货物系统的风险评价精准度。作者以粒子群算法为研究对象,提出基于相交基学习的重构型精英集粒子群算法,作为核心训练神经网络,从而实现并且优化出入境货物系统中对出入境货物的风险评价、预测,以满足出入境货物系统中对出入境货物抽检的智能化需求。该文提出三种改进策略,包括：通过对生物种群中精英群体的数学仿生,给出粒子群算法关键参数个体学习、社会分享系数的动态调整策略；为克服算法因陷入局部最优而停滞的问题引入重构概念；通过引入相交基学习算法用于粒子群算法的初始化,以防止因随机初始化而导致的不稳定问题。该文选取了不同学习拓扑结构的三种改进粒子群算法,与所提出基于相交基学习的重构型精英集粒子群算法进行比对研究：包括其收敛速度、鲁棒性、稳定区域。最后对出入境货物风险问题及系统框架进行阐述并利用基于相交基学习的重构型精英集粒子群算法训练神经网络,与其他三种神经网络进行系统比对试验,证实其在出入境货物风险评价中的可行性及更优精准度。
[Abstract]:With the rapid development of economy and trade, the large increase of trade volume brings challenges to the inspection of inbound and outbound goods. And can not effectively identify substandard products. Considering the higher speed and higher precision required by the sampling inspection of cloth control. This paper presents a risk assessment system for inbound and outbound cargo to realize intelligent random check. The author proposes a reconfigurable elite particle swarm optimization algorithm based on intersecting basis learning to train neural network to optimize the accuracy of risk evaluation in the cargo system. Particle Swarm Optimization (PSO) is the object of study. A reconfigurable elite particle swarm optimization algorithm based on intersecting basis learning is proposed to train neural network as the core to realize and optimize the risk evaluation and prediction of entry and exit cargo in the system of entry and exit cargo. In order to meet the intelligent demand of the immigration and entry cargo system, this paper puts forward three improved strategies, including the mathematical bionics of the elite group in the biological population. The key parameters of particle swarm optimization (PSO), individual learning and social sharing coefficient, are dynamically adjusted. The concept of reconstruction is introduced to overcome the stagnation of the algorithm due to local optimization. In order to prevent the instability caused by random initialization, three kinds of improved particle swarm optimization (PSO) algorithms with different learning topologies are selected by introducing intersecting basis learning algorithm to initialize particle swarm optimization (PSO). Compared with the proposed reconfigurable elite particle swarm optimization algorithm based on intersecting basis learning, including its convergence speed and robustness. Finally, the risk problem and system framework of inbound and outbound cargo are expounded, and the neural network is trained by the reconstructed elite particle swarm optimization algorithm based on intersecting basis learning. Compared with the other three neural networks, the feasibility and better accuracy of the neural network in the risk assessment of inbound and outbound cargo are verified.
【学位授予单位】：北京林业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F752.5;TP18

【相似文献】