点过程序列数据的建模与分类方法研究

发布时间：2018-07-15 18:02

【摘要】：在临床研究中，以删失数据更为常见。由于死亡时间的不确定，只能将其看做是一个大的范围，又由于点过程理论的前提是假设点事件发生在一个很小的范围之内，因而删失数据不能将其看做为一个点事件。对删失数据的点过程化和生存数据建模与分析是一个重要的课题。本文提出了基于点过程理论的生存函数非参数估计，并将该方法应用到模拟数据和真实乳腺癌患者数据的生存曲线的非参数估计，并同传统SC方法进行比较。熵广泛的应用到两组脑电信号和心电信号检测与分类中，并得到了较好的效果。但熵只提取了混沌信号中的复杂程度，忽略了两种信号中最大幅值的区别，因此在两个复杂程度相似但幅值有很大差别的信号中，，熵的分类效果就大大降低。将信号中最大幅值信息添加到样本熵的方法中，本文提出一种基于多元点过程熵的信号分类方法。将该方法应用到波恩大学癫痫病中心公开脑电数据，其分类结果同只基于原始脑电数据的传统多元多尺度熵的分类结果进行对比，结果比较表明本文提出的方法的精确度高于传统多元多尺度熵。最后向波恩癫痫脑电数据加入不同等级的白噪声，并同时应用本文方法和传统多元多尺度熵进行分类，结果比较说明：多元点过程熵在对干扰性大的信号分析中较传统方法有更高的精度。
[Abstract]:Deletion of data is more common in clinical studies. Because of the uncertainty of the time of death, it can only be regarded as a large range, and because the premise of the point process theory is that the point event is assumed to occur in a very small range, the censored data cannot be regarded as a point event. It is an important task to model and analyze the point process and survival data of censored data. In this paper, the nonparametric estimation of survival function based on point process theory is proposed. The proposed method is applied to the nonparametric estimation of survival curve between simulated data and real breast cancer data, and compared with the traditional SC method. Entropy is widely used in the detection and classification of two groups of EEG and ECG signals. However, entropy can only extract the complexity of chaotic signals and ignore the difference of the largest values in the two signals. Therefore, in two signals with similar complexity but great difference in amplitude, the classification effect of entropy is greatly reduced. In this paper, a signal classification method based on multivariate point process entropy is proposed by adding the maximum value information to the sample entropy. The method was applied to the published EEG data from the epilepsy center of the University of Bonn. The classification results were compared with the traditional multivariate multi-scale entropy classification results based only on the original EEG data. The results show that the accuracy of the proposed method is higher than that of the traditional multivariate multi-scale entropy. Finally, different levels of white noise are added to the epileptic EEG data in Bonn, and at the same time, the method of this paper and the traditional multi-scale entropy are used to classify the EEG data. The results show that the multivariate point process entropy has higher accuracy than the traditional method in signal analysis with large interference.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.6;R742.1

【参考文献】