基于CEEMDAN-ELM的短期血糖预测模型研究

发布时间：2018-01-14 04:31

  本文关键词：基于CEEMDAN-ELM的短期血糖预测模型研究　出处：《郑州大学》2017年硕士论文　论文类型：学位论文

  更多相关文章： 血糖预测 极限学习机 CEEMDAN 低血糖预警 CGM

【摘要】：由于生活方式的改变,糖尿病患病群体呈年轻化趋势,已成为严重影响人类健康的疾病。血糖预测是人工胰脏血糖闭环控制的关键,可通过调控胰岛素注射剂量和时间强化糖尿病患者体内的血糖控制并降低并发症的发生,为医生和患者进行血糖控制提供数据支持。因此提高血糖预测精度、增加预测时间具有十分重要意义。血糖预测是根据人体内的历史血糖值来预测未来一段时间血糖浓度的变化趋势。本文在研究已有的血糖预测技术基础上,提出了一种基于自适应白噪声完整聚合经验模态分解-极限学习机(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Extreme Learning Machine,CEEMDAN-ELM)的短期血糖预测模型。为提高血糖预测的精确度,模型首先运用信号分析技术,利用CEEMDAN方法对糖尿病患者血糖值时间序列进行平稳化处理,逐级分解患者血糖数据中存在的不同尺度下的波动或变化趋势以降低血糖值时间序列的非线性和非平稳性,获得一组含有不同频段特征的血糖分量;然后对各血糖分量分别利用ELM进行预测;最后融合各血糖分量的预测结果,获得糖尿病患者最终的血糖预测值。论文在CEEMDAN-ELM短期血糖预测模型基础上进行了低血糖预警研究,设计了一种新的低血糖预警方法。本文利用60例Ⅱ型糖尿病患者的血糖数据进行血糖预测实验,并采用克拉克网格分析法、配对t检验等对预测模型的性能进行了检验,采用虚警率和漏警率来评估低血糖预警效果。结果表明:与ELM模型和EMD-ELM模型相比,CEEMDAN-ELM短期血糖预测模型提前45min的血糖预测达到了较高的精确度(60例患者的平均RMSE=0.2046,MAPE=2.0855%);提前45min的患者血糖预测值基本落在了Clarke网格误差图的A区域。在CEEMDAN-ELM短期血糖预测模型的基础上提出的低血糖预警方法使得26例具有低血糖事件的糖尿病患者的虚警率为0.77%,漏警率为8.71%。CEEMDAN-ELM血糖预测模型提高了预测精度,延长了预测时间,对提高糖尿病的治疗效果具有重要意义。
[Abstract]:Because of the change of life style, the diabetic patients tend to be younger, which has become a serious disease affecting human health. Blood glucose prediction is the key to the closed loop control of blood glucose in artificial pancreas. It can enhance the blood glucose control and reduce the incidence of complications in diabetic patients by regulating the dose and time of insulin injection, which can provide data support for doctors and patients to carry out blood glucose control. Therefore, improve the accuracy of blood glucose prediction. It is very important to increase the predicted time. The prediction of blood sugar is based on the historical blood sugar value in human body to predict the trend of blood glucose concentration in the future. An adaptive white noise complete aggregation empirical mode decomposition-extreme learning machine (LLMs) based on adaptive white noise is proposed in this paper. Complete Ensemble Empirical Mode Decomposition with Adaptive. Noise-Extreme Learning Machine. CEEMDAN-ELM) short-term blood glucose prediction model. In order to improve the accuracy of blood glucose prediction, the model first used signal analysis technology. The CEEMDAN method was used to stabilize the time series of blood glucose in diabetic patients. In order to reduce the nonlinearity and nonstationarity of the time series of blood glucose values, a group of blood glucose components with different frequency characteristics was obtained by decomposing the fluctuation or changing trend of different scales in the blood glucose data step by step. Then the blood glucose components were predicted by ELM. Finally, the final blood glucose prediction value of diabetic patients was obtained by combining the predicted results of each component of blood sugar. Based on the CEEMDAN-ELM short-term blood glucose prediction model, hypoglycemia prediction was studied in this paper. A new early warning method for hypoglycemia was designed in this paper. The blood glucose prediction experiment was carried out by using the blood glucose data of 60 patients with type 鈪，

本文编号：1422072