摘要
肥胖人群是特殊人群,通过实时检测血压,可以预防高血压,大大降低由高血压引起的各种并发症。为了解决现有算法对肥胖人群血压测量准确度不高的问题,提出一种基于Stacking集成机器学习的血压计算模型,该模型将K近邻、极端随机树、lightGBM回归模型作为初级学习器,线性回归模型作为次级学习器。通过使用长桑技术设备提取并建立PPG特征值数据集,将数据集分成非肥胖人群(BMI<25)和肥胖人群(BMI>25),并把身体质量指数(BMI)作为新的特征参数加入到模型进行训练和测试。实验结果:在非肥胖人群中,Stacking模型测量收缩压和舒张压的评价指标RMSE/MAE分别为6.611/5.410和4.368/3.242;在肥胖人群中,Stacking模型测量收缩压(SBP)和舒张压(DBP)的评价指标RMSE/MAE分别为6.394/4.979和4.350/3.233。实验结果表明,该Stacking模型对肥胖人群的血压计算精度明显高于非肥胖人群,且测量结果符合AAMI国际电子血压计标准(RMSE<8 mmHg,MAE<5 mmHg),提高了肥胖人群血压测量的准确度,可以应用到生物医学领域。
Obese population are special people.Through real-time detection of blood pressure,hypertension can be prevented and various complications caused by hypertension can be greatly reduced.In order to solve the problem of low accuracy of existing algorithms for blood pressure measurement of obese population,a blood pressure calculation model was proposed based on stacking ensemble machine learning.The K-nearest neighbor,extreme random tree and light GBM regression model were taken as primary learners and linear regression model was taken as secondary learners.The PPG eigenvalue data set was extracted and established by using changsang technology and equipment and the data set was divided into non-obese population(BMI<25)and obese population(BMI>25)and the body mass index(BMI)was added to the model as a new characteristic parameter for training and testing.The results are as follows:in the non-obese population,the RMSE/MAE of systolic and diastolic blood pressure measured by stacking model is 6.611/5.410 and 4.368/3.242 respectively.In obese population,the RMSE/MAE of systolic blood pressure(SBP)and diastolic blood pressure(DBP)measured by stacking model is 6.394/4.979 and 4.350/3.233respectively.The experimental results show that the accuracy of the stacking model used is significantly higher than that of the non-obese people,and the measurement results meet the AAMI international electronic sphygmomanometer standard(RMSE<8 mmHg,MAE<5 mmHg),which improves the accuracy of blood pressure measurement in the obese population and can be applied to the biomedical field.
作者
李清福
赵宇波
赵景波
蒋泽宇
Li Qingfu;Zhao Yubo;Zhao Jingbo;Jiang Zeyu(School of Information and Control Engineering,Qingdao University of Technology,Qingdao,Shandong 266520,China;Shandong Institutes of Industrial Technology(Qingdao),Qingdao,Shandong 266101,China)
出处
《机电工程技术》
2023年第2期123-129,共7页
Mechanical & Electrical Engineering Technology
基金
国家重点研发计划项目新型研发机构创新服务平台技术研发与应用(编号:2021YFF0901104)。
