带蜂群策略的粒子群算法训练人工神经网络研究  

Research on particle swarm optimization algorithm with strategies of bee colony for training artificial neural network

作　　者：林晓宇[1] 钟一文[1] 

机构地区：[1]福建农林大学计算机与信息学院,福建福州350002

出　　处：《计算机工程与设计》2011年第10期3514-3517,共4页Computer Engineering and Design

基　　金：福建省自然科学基金项目(2008J0316);福建省教育厅科技基金项目(JA10118)

摘　　要：提出一种带蜂群策略的粒子群优化算法,并将算法应用于神经网络训练。将蜂群优化算法中引领蜂和观察蜂的收益评价与贪婪选择策略以及侦察蜂的探索新解策略引入到粒子群优化算法中。粒子在飞行时,按维度对粒子速度和位置进行更新,根据对收益的评价,只接收能够提高解适应值的位置,从而加快了收敛速度;如果粒子多次迭代均无法改进解,则在解空间中随机搜索新的位置,增强算法跳出局部极值的能力。在求解异或问题、奇偶校验和编码解码问题的神经网络上进行了仿真,结果表明,该算法优于BP算法、粒子群优化算法和蜂群优化算法。A particle swarm optimization algorithm with strategies of bee colony is proposed and used for the training of artificial neuralnetwork. The proposed algorithm uses the solution evaluation and greedy selection strategies of employed bee and onlookers, and usesthe searching new solution strategy of scouts. Particle updates and evaluates its posltlon dimension by dimension wnen it ny m me searcn space, and only accepts those new position which can improve the solution. This greedy strategy improves its convergence speed effectively. In order to help the algorithm to escape from local minimum, particle will search new position randomly when its solution cannot be improved for specified iterations. The simulation experiments, which were carried on the training of neural network for XOR, Nbit parity and encoder-decoder problems, show that the proposed algorithm had better performance than the BP algorithm, particle swarm optimization algorithm and bee colony optimization algorithm.

关 键 词：粒子群优化 蜂群优化 人工神经网络 异或问题 奇偶校验 编码解码问题 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]