【作者】 蒋贤海；

【导师】 谢存禧；

【作者基本信息】 华南理工大学 ， 机械设计及理论， 2011， 博士

【摘要】 随着人民生活水平的提高,人类平均寿命普遍延长,人口老龄化日益明显。提高老年人健康生活质量是今后卫生工作重点。远程健康监护是我国近十年来蓬勃发展起来的课题,它与医学、信息科学、计算机、网络与通信等密切相关,是社会发展的必然需求。国内外学者已对监护系统做了大量的工作,并且已经提出了许多监护信息处理方法。然而在实际应用中,传统的监护信息处理方法,大多都存在各自的缺陷或受到适用条件的限制。随着社会的进步,希望所开发的监护系统具有更准确更综合的功能。因此,研究远程健康监护系统监护信息分析处理方法具有重要的现实意义和推广价值。本文在对监护信息的特点及相关的技术进行了分析,以及国内外研究现状进行分析研究的基础上,提出了远程健康监护系统中监护信息数据分析方法。针对监护过程所采集的海量数据,采用小波分析、神经网络、支持向量机等数据挖掘技术对用户的多生理参数进行分析处理,消除冗余数据,对影响分析的异常数据进行识别处理,采用预报技术实现监护信息的预报。本文的主要工作和贡献有以下几方面：1.提出了对监护信息去除干扰信号的处理方法。由于监护信号常受到如噪声、伪迹的干扰,造成监护信号参数的估值错误,从而导致监护仪错误报警。由于小波变换能有效的抑制监护对象参数中所含的噪声和干扰信号,对微弱的噪声等干扰信息,本文提出基于改进阀值的小波变换对其进行处理；对噪声等干扰信号较强的,采用小波变换和Hampel相结合的方法降噪,并分别与采用传统小波阀值去噪方法进行对比实验,结果表明所提出的降噪方法能达到较好的效果,以提高诊断报警的准确性和有效性。2.提出了远程健康监护系统的监护信息异常值识别处理方法。对多种处理异常值的方法进行研究,针对监护信息异常值识别问题,提出了利用改进统计量的假设检验方法与基于Kalman滤波的AR(p)模型的缺失值重构的监护信息异常值识别处理方法,首先采用改进统计量的假设检验方法识别出监护数据中可能的异常值,并将可能的异常点进行直接删除处理。然后视已删除的异常点为缺失值,采用基于Kalman滤波算法的AR(p)模型对缺失值进行预测,以该时刻的预测值添补缺失值,为下一步的准确预报做准备。3.为实现监护系统的报警功能,建立了基于小波分解与最小二乘支持向量回归机的监护信息综合预报模型。首先对监护信息进行小波分解,得到各分解层序列；然后对各分解层序列应用最小二乘支持向量回归机对各序列进行回归预测,最后对各序列预测结果进行叠加,实现监护信息的预报。同时采用PSO寻参方法对最小二乘支持向量机进行参数寻优,对提出的方法进行了实验验证,并与其他预报方法进行对比,证明所提出的方法能取得较好的预报结果。4.研制了智能远程健康监护系统实验平台。给出了智能远程健康监护系统的总体结构、硬件平台及系统的软件设计,利用Windows NT操作系统,SQL Server 2000数据库,Windows visual studio.net 2005为工具开发了智能远程健康监护系统软件,实验结果表明健康监护系统能有效的降低噪声干扰,对异常值进行识别处理并对监护信息进行预报,进一步证实了提出的方法的有效性。更多还原

【Abstract】 With the improvement of the people’s lives, the expectation of life is longing, and the old people is becoming more and more. To meet with the health of old people and improve the living quality, Community medical service should be strengthened aiming at the focus on sanitation of old people’s healthy life. In China, tele-monitoring of health has been developed in recent decade, it is the inevitable result with the development of society, and it is closely related with medicine, information science, computer, network and communication. Many methods of monitoring information process have been put forward as a lot of work for monitoring system has been done by scholars both at home and abroad. However, the traditional processing methods have its defects or suitable conditions in the actual application. The hope of monitoring system is more accurate and more comprehensive with the progress of the society. Therefore, the study on the tele-monitoring system and the processing methods of health information has important realistic meaning and extension value.The analyzing methods of monitoring information have been put forward on the basis of analysis of the characteristics of monitoring information and its related technologies, and the recent studies. According to the mass monitoring data, the technologies of data mining for physical parameters, such as wavelet analysis, support vector machines and neural networks, have been used to eliminate redundant data, identify abnormal data in physical parameters which will influence diagnosis, and forecast techniques have been used to realize forecast of monitoring information.The main work and contributions of the paper are as follows.1. The signal processing methods for monitoring information have been brought forward. As the monitoring signals are disturbed by noise, artifact and data loss, the error valuations of signal parameter, and it leads to the error alarm of guardianship. Because of wavelet transform can restrain jamming signal and noise contained in monitoring information effectively, the method of wavelet transform based on improved threshold is put forward to handle faint noise interference. While the combined method of wavelet transform and Hampel is adopted to strong noise signal. The experiment results show that the proposed denoising method can achieve good results compared with traditional wavelet threshold, and it can improve the accuracy and validity of alarm.2. The methods of identifying abnormal values of monitoring information have been raised. According to the problem of identifying abnormal monitoring information, the abnormal recognition method of monitoring information is put forward, which is combined hypothesis based on improved statistics with AR(p) based on Kalman filter, after studying on many other deposing method of abnormal. First, the abnormal monitoring data has been identified by means of improving hypothesis, and it will be delete directly. Then the deleted points are took as missing value, and the abnormal values are be predicted with AR(p) model based on Kalman filtering algorithm, and we fill the abnormal values with the predicted values, which is the preparation for next prediction.3. To realize the alarm of monitoring system, the integrated forecast model of monitoring information based on wavelet decomposition and least squares supports vector regression are established. Firstly, obtaining each decomposed layers sequence of monitoring information with wavelet decomposition. Secondly, the each decomposed layers sequence are predicted by using the least square support vector regression machine. Finally, realizing the prediction of monitoring information through composing the forecast results of each sequence. Meanwhile, the LS-SVMs are optimized by using PSO, and the related experiments are conducted and the results show the integrated method can obtain better forecasting results than other forecast methods.4. The article develops the experiment platform of intelligent health monitoring system. System structure, hardware and software architecture are designed. Healthy monitoring system is programmed on the operation system Windows NT. SQL Server 2000 is used as database, and Windows visual studio.net 2005 is adopted as development tool. Experiment results show that health monitoring system has the function of reducing noise effectively, identification of unusual values and prediction. The efficiency of proposed methods are intensified by experiment results.更多还原