论文部分内容阅读
利用过程神经元网络对动态时变信号过程模态特征的自适应提取能力,结合半监督学习算法,提出了一种基于半监督学习的网络结构自适应的竞争型过程神经元网络用于示功图识别.网络采用广义离散Fréchet距离作为动态样本间距离的测度,然后直接以离散化的载荷和位移时间序列作为网络输入,在样本标记信息的约束下,采用奖励—惩罚更新规则,根据网络学习目标函数,动态重构竞争层节点,消除网络对初始聚类数的依赖,实现样本的有效聚类.仿真实验结果验证了模型和算法的有效性.
Based on the semi-supervised learning algorithm, a neural network based on semi-supervised learning with self-adaptive network structure is proposed for the process of self-adaptive extraction of modal features of dynamic time-varying signal process. Graph recognition.Network uses the generalized discrete Fréchet distance as a measure of the distance between dynamic samples and then uses the discretized load and displacement time series directly as the input of the network.Under the constraints of the sample mark information, The objective function is to dynamically reconstruct the nodes in the competition layer and eliminate the dependence of the network on the initial clustering number, so as to effectively cluster the samples.The simulation results verify the effectiveness of the model and the algorithm.