基于人工神经网络的动态过程聚类研究

Study on dynamic process clustering based on artificial neural network

为了有效解决传统人工神经网络对于时变函数的聚类问题,以及提高在大样本下网络的学习和泛化能力,提出了基于离散余弦变换的传统人工神经网络动态过程聚类方法。通过离散余弦变换将样本函数降维映射到由对应余弦参数所张成的模式特征空间,满足了传统人工神经网络对输入样本的要求,使传统人工神经网络实现动态过程的聚类成为可能。给出了实现算法,分析了计算复杂度,并使用基本竞争型人工神经网络对特征样本向量进行聚类,实验结果表明该方法是正确、有效的。与过程人工神经网络相比,该方法具有运算简单、物理意义明确等优点。

Traditional artificial neural network can not solve the clustering problem of the time change function effectively,and can not improve the net＇s capability of learning and generalization in the case of massive samples.We propose a new method based on Discrete Cosine Transform （DCT） for artificial neural network to try to solve these problems.First the original signal sequence is mapped into pattern space by making use of DCT,and then we use the basic competition neural network to cluster these vectors.The result shows that the method is effective and has the advantages of simple operation and specific physics meaning compared with the procedure artificial neural network.