摘要
针对时序背景下的聚类问题,提出一种基于小波和改进自组织过程神经网络的时序聚类方法,首先应用小波变换对原时序数据进行小波分解,在保留相关聚类特征的原则下,对信号进行重构;然后将重构信号拟合为时变函数作为过程神经网络的输入,应用改进的竞争算法训练自组织过程神经网络,利用过程神经网络输入为时变函数的特点,将经过小波处理后的时序信号特征充分考虑到聚类分析中,网络提取输入函数隐含的过程式模式特征并进行自组织,给出了改进的竞争学习算法;最后应用UCI数据集聚类结果表明,该方法在聚类正确率、网络运行时间和收敛速度上均有提高,同时在聚类质量、聚类速度方面表现出良好性能,能有效地应用于时序聚类。
For time series clustering problem,a method based on wavelet and improved self-organization process neural networks（PNN） was proposed.First,original time series data was decomposed by wavelet.Under the principle of reserving clustering characteristics,the signal was reconstructed.And then reconstructed signal fitted into time-varying functions was used as PNN＇s input.Self-organization PNN was trained by improved competition algorithm.Making use of time-varying input characteristic of PNN,the timing signal characteristics processed by wavelet has been considered adequately in clustering analysis.Network extracts implicit process mode characteristics of function to self organize.The improved competition learning algorithm was given.Finally,clustering result of UCI datasets shows that the proposed approach has an improvement in clustering accuracy,network runtime and convergence speed,at the same time shows good performance in clustering accuracy and clustering speed,can be applied to timing clustering effectively.
出处
《电机与控制学报》
EI
CSCD
北大核心
2011年第12期78-82,共5页
Electric Machines and Control
基金
黑龙江省科技攻关计划项目(GC05A118)
哈尔滨市科技创新人才研究专项资金项目(2008RFQXG072)
关键词
时间序列
聚类分析
自组织过程神经网络
小波
time series
clustering analysis
self-organization process neural networks
wavelet
