摘要
针对传统BP-WNN和基本PSO-WNN算法收敛速度慢和泛化性能低的缺陷,在应用李雅普诺夫理论分析得到单个粒子稳定收敛的参数取值条件基础上,提出一种粒子群改进算法,并利用该算法来训练小波神经网络权值,以此构建一种高效的粒子群小波神经网络分类器。通过Iris标准分类数据集进行测试,结果表明所提出的改进算法与BP-WNN,PSO-WNN等经典算法相比,网络更易于全局收敛,迭代次数少、函数逼近误差小、分类精度高。将该分类器应用于非线性辨识和固井质量评价中,均取得了不错的效果,表明该分类器泛化能力强,具有良好的使用价值和应用前景。
In view of the defects that traditional BP-WNN algorithm and basic PSO-WNN algorithm have slow convergence speed and low generalization capability, Lyapunov stability theory is used to obtain the convergence condition of single particle. Based on that, a new strategy is introduced to improve the performance of PSO. Then, an improved PSO algorithm is used to train the parameters of WNN, and a high efficient classifier based on the improved PSO-WNN is created. The proposed algorithm is compared with BP-WNN and basic PSO-WNN algorithms through analyzing the Iris benchmark data set. Simulation results confirm that the performances of the new algorithm, such as global convergence, iterative number, error of function approximation and classification accuracy rate, are highly improved. Finally, the new classifier is applied to nonlinear model identification and cement bond quality evaluation. Experiment results show that the classifier has strong generalization capability, and has high application value and good prospect.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2010年第10期2203-2209,共7页
Chinese Journal of Scientific Instrument
基金
国家高技术研究发展863计划(2006AA06Z222)资助项目
关键词
收敛速度
粒子群算法
小波神经网络
分类器
泛化
convergence speed
particle swarm optimizer algorithm
wavelet neural network
classifier
generalization
