基于粗糙集和SVM的体域网健康评估方法研究

发布时间：2018-01-02 13:39

本文关键词：基于粗糙集和SVM的体域网健康评估方法研究　出处：《浙江师范大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着无线传感器的广泛应用,无线人体区域网络(简称体域网)将极大地推动医院智能监护体系的发展。为了让高龄、独居老人的健康状态得到很好地监测和保护,体域网健康估计方法的研究将具有重大的现实意义。体域网的健康评估就是专门针对人体的生理指标数据进行评估的技术。但是采集的患者的生理参数数据大部分时间段内是稳定不变,因此只需要分析异常时间阶段的生理指标数据来进行健康评估。目前,一系列传统的算法是通过异常检测的方式来获取训练的数据集的方法。它们利用时间序列的趋势分析、平稳的时间序列的数学模型等方法预测下一刻会出现异常的时刻,并通过计算预测值与实际值残差方差的方式设置异常概率的阈值和发生连续异常的时间阈值来检测异常。而且传统的用于健康评估BP神经网络模型虽具有较好的效果,预测准确度也较高,但参数往往依赖经验值,网络中存在较大的冗余性,需要较多的迭代次数,容易造成过度学习等缺点。为了更好的解决在健康评估所面临的以上问题,本文在改进健康评估算法方面做了一些研究工作,具体的内容如下所述：(1)目前,常见的健康评估算法没有考虑到各个样本中的稳定的生理指标值可被其余同类样本一一对应表征,我们可以把异常时间段的多个正常数据用一个同类数据来表征,这样可以大大降低数据的冗余。同时我们考虑到数据在内存中二进制存储的特性,采用优化链表结构的属性约简的粗糙集算法来实现信息存储的内存优化。(2)在用于健康评估的机器学习算法中,针对BP (Back Propagation)神经网络模型隐含层的层数和个数是经验值,需要通过反复实验确定,网络的学习和记忆能力不稳定性等缺点,本文将SVM (Support Vector Machine)的学习模型引入到评估算法中。SVM健康评估模型不仅能保证分类的准确性,而且能降低学习的VC维(Vapnik-Chervonenkis Dimensbn)度,得到全局的最优解,如此预测样本数据的泛化能力优于BP神经网络方法。同时线性组合多项式核函数、径向基核函数来改进SVM的核函数,这样就可以大大简化机器学习模型的迭代次数,提升健康评估模型的性能。最后提出一种基于改进SVM健康评估算法。
[Abstract]:With the wide application of wireless sensors, wireless human body area network (WSNs) will greatly promote the development of hospital intelligent monitoring system. The health status of the elderly living alone is well monitored and protected. The research on the health estimation method of body area network will be of great practical significance. The health evaluation of body area network is a special technology to evaluate the physiological index data of human body. But most of the patient's physiological parameter data are collected. Steady over time. Therefore, we only need to analyze the physiological index data of abnormal time stage to carry on the health evaluation. At present. A series of traditional algorithms are used to obtain the trained data set by anomaly detection. They use the trend analysis of time series. Stationary time series mathematical models and other methods to predict the next moment will be abnormal. The outlier is detected by calculating the variance of residual error between the predicted value and the actual value by setting the threshold of abnormal probability and the time threshold of occurrence of continuous anomaly, and the traditional BP neural network model for health assessment is better than the traditional BP neural network model. The effect. The prediction accuracy is also high, but the parameters often depend on the experience value, and the network has more redundancy, so it needs more iterations. In order to better solve the above problems in health assessment, this paper has done some research work in improving the algorithm of health assessment. At present, common health assessment algorithms do not take into account that the stable physiological index values in each sample can be represented one-to-one by the other similar samples. We can represent more than one normal data in an abnormal time period with one of the same kind of data, which can greatly reduce the redundancy of the data, and we also take into account the data in memory binary storage characteristics. A rough set algorithm based on attribute reduction of optimized linked list structure is used to realize memory optimization of information storage. It is used in machine learning algorithm for health assessment. The number and number of hidden layers in BP back propagation neural network model are empirical values, which need to be determined through repeated experiments. The learning and memory ability of network is unstable and so on. In this paper, the learning model of SVM support Vector Machine is introduced into the evaluation algorithm. The health evaluation model of SVM can not only guarantee the accuracy of classification. Moreover, the Vapnik-Chervonenkis Dimensbnn degree of learning can be reduced, and the global optimal solution can be obtained. The generalization ability of predicting sample data is better than that of BP neural network, and the kernel function of SVM is improved by linear combination polynomial kernel function and radial basis kernel function. In this way, the iterations of the machine learning model can be greatly simplified and the performance of the health assessment model can be improved. Finally, an improved SVM health assessment algorithm is proposed.
【学位授予单位】：浙江师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN92;TP18

【相似文献】