采用改进的布谷鸟算法优化极限学习机  

Optimized Extreme Learning Machine with Improved Cuckoo Algorithm

作　　者：赵坤 覃锡忠[1] 贾振红[1] ZHAO Kun;QIN Xi-zhong;JIA Zhen-hong(College of Information Science and Engineering,Xinjiang University,Xinjiang,Urumqi 830000,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830000

出　　处：《计算机仿真》2018年第11期236-241,共6页Computer Simulation

基　　金：中国移动通信集团新疆有限公司研究发展基金项目(XTM2013-2788)

摘　　要：针对极限学习机在处理非线性问题时,网络结构难以确定,将导致算法精度低、稳定性差的问题,提出利用改进的布谷鸟搜索算法优化极限学习机的算法。引入差分进化算法的变异策略对布谷鸟搜索算法进行改进,从而减少了迭代次数、增强了全局搜索能力;然后采用改进后的CS算法自适应的选择极限学习机的隐层神经元的个数及其所对应的输入权值和阈值,以提高模型的精度和稳定性。对不同数据的时间序列预测的仿真结果表明,提出的算法与其它算法相比,在收敛速度、预测精度和稳定性方面都有明显的提高。把提出的算法应用到多步预测中,进一步验证了提出的算法的有效性。In order to deal with the nonlinearity in Extreme Learning Machine（ ELM）,an algorithm based on Improved Cuckoo Search（ ICS） algorithm to optimize ELM（ ICS-ELM） was proposed. The mutation strategy of the Differential Evolution（ DE） algorithm was introduced to improve the Cuckoo Search（ CS） algorithm,which reduces the number of iterations and enhances the global search ability. And in order to improve the accuracy and stability of the model,the improved CS algorithm was used to optimize the number of hidden layer neurons of ELM and the corresponding input weights and thresholds were selected adaptively. The simulation results of time series prediction of different data show that the proposed algorithm has obvious improvement in convergence speed,prediction accuracy and stability compared with other algorithms. The proposed algorithm is applied to multi-step prediction,which further validates the validity of the proposed algorithm.

关 键 词：布谷鸟搜索 极限学习机 自适应 稳定性 多步预测 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]