基于ARIMA和BPNN的组合预测模型在血糖预测中的应用

发布时间：2018-01-30 18:37

本文关键词： 血糖预测小波去噪 ARIMA BP神经网络组合预测　出处：《郑州大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着糖尿病患者数量的不断增多,糖尿病对人类健康的危害日趋增加,而稳定血糖是糖尿病患者临床治疗的主要目的,如果能提前预测出患者的血糖浓度,那么医生和患者就能在高血糖或者低血糖事件发生之前采取措施来稳定血糖,这将极大减小糖尿病对患者造成的伤害。建立一个精确度比较高的血糖预测模型,为医生和糖尿病患者提供指导,具有很好的应用价值。目前,关于人体血糖预测技术的研究大体有两个方向:一个方向是只利用患者的历史血糖值,不考虑影响患者血糖动态变化的外部因素(饮食、药物注射、运动等),追求简单和高效,但不够精准;另外一个方向不仅利用糖尿病患者的历史血糖值,而且结合人体的生理模型和大量的病理学、生理学的知识,追求准确和精准,算法复杂,有一定延迟。本文深入研究了影响人体血糖变化的关键因素和血糖预测所面临的问题,在比较了现有的血糖预测技术的基础上,探讨基于ARIMA和BPNN的组合预测模型对患者血糖未来值分析和预测的可行性。采用ARIMA对糖尿病患者的历史血糖值进行分析,找出患者血糖变化的线性规律,利用BPNN捕获外部因素对人体血糖的影响,并对输入值、误差项等进行学习和拟合。最后将ARIMA计算出的预测值与BP算法得到的修正值进行组合,得到准确的结果。同时,针对饮食或药物注射在短时间内对人体血糖波动的突发影响,设定开始影响的点为奇异点,提出一种奇异点发现和处理算法,在人体血糖受外部干扰发生不规律变化时自动调整未来一段时间内的预测值,保证组合预测模型的精度和准确度。采用河南省人民医院内分泌科所提供的糖尿病患者血糖数据对所提出来的基于ARIMA和BPNN的组合预测模型及奇异点发现和处理算法进行验证。结果表明,相比ARIMA预测,所提出来的组合预测模型具有更好的预测效果,可以给医生或者糖尿病患者提供临床上的指导;所提出的奇异点发现和处理算法,在人体血糖受外部干扰发生急剧变化时能自动调整未来一段时间内的预测值,能保证组合预测模型的预测精度和准确度。
[Abstract]:With the increasing number of patients with diabetes, diabetes is increasingly harmful to human health, and stable blood sugar is the main purpose of clinical treatment of patients with diabetes, if we can predict the concentration of blood sugar in advance. Then doctors and patients can take steps to stabilize blood sugar before hyperglycemia or hypoglycemia occurs, which will greatly reduce the damage caused by diabetes. To provide guidance for doctors and patients with diabetes, has a good application value. At present, there are two directions in the study of blood glucose prediction technology: one direction is to use only the patient's historical blood sugar value. Regardless of the external factors (diet, drug injection, exercise, etc.) that affect the dynamic changes of blood glucose, the pursuit of simplicity and efficiency, but not accurate; The other direction not only uses the historical blood sugar value of diabetic patients, but also combines the physiological model of the human body and a lot of pathology, physiological knowledge, the pursuit of accuracy and precision, complex algorithm. In this paper, the key factors affecting the changes of blood glucose and the problems of blood glucose prediction are studied in depth, based on the comparison of existing blood glucose prediction techniques. To explore the feasibility of the combination prediction model based on ARIMA and BPNN to analyze and predict the future value of blood glucose in patients with diabetes mellitus. ARIMA was used to analyze the historical blood glucose value of patients with diabetes mellitus. To find out the linear rule of blood glucose change, use BPNN to capture the influence of external factors on human blood sugar, and to input value. Finally, the predicted value of ARIMA is combined with the revised value of BP algorithm to get accurate results. Aiming at the sudden effect of diet or drug injection on blood glucose fluctuation in a short period of time, the singularity point is set as the starting point, and a singular point detection and processing algorithm is proposed. Automatically adjust the predicted value for a period of time when the body's blood sugar changes irregularly by external interference. To ensure the accuracy and accuracy of the combined prediction model. The combined prediction model based on ARIMA and BPNN was proposed by using the blood glucose data of diabetic patients provided by the Endocrinology Department of Henan Provincial people's Hospital. The algorithm of point discovery and processing is verified. The results show that. Compared with ARIMA prediction, the proposed combined prediction model has better predictive effect and can provide clinical guidance to doctors or patients with diabetes. The proposed singular point detection and processing algorithm can automatically adjust the prediction value in the future when the blood sugar changes sharply by external interference, which can ensure the prediction accuracy and accuracy of the combined prediction model.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R587.1;TP18

【参考文献】