一种实值深度置信网络在人体血压预测中的应用

APPLICATION OF A REAL-VALUED DEEP BELIEF NETWORK IN PREDICTION OF HUMAN BLOOD PRESSURE

针对现有血压预测方法难以表征药物治疗、生活方式干预等降压措施对于血压长期且复杂的影响,提出一种基于实值深度置信网络(Real-valued Deep Belief Nets,RDBN)的人体血压预测模型,用于预测高血压患者及高危人群采取降压措施一个随访周期后的血压变化情况。该模型通过双层高斯结构的受限波尔兹曼机单元(Gaussian-Gaussian Restricted Boltzmann Machine,GG-RBM)堆叠形成的深层网络自动提取作用于未来血压的影响因素,挖掘血压与其影响因素之间复杂的非线性关系。基于自适应距估计(Adaptive Moment Estimation,Adam)算法加速参数空间的寻优过程。实验结果表明,对于患者未来血压的预测,基于RDBN的血压预测模型与传统预测方法相比,具有更高的预测精度。

The blood pressure prediction methods are difficulty in characterizing the long-term and complex effects of antihypertensive measures such as medications and lifestyle interventions.To solve this problem,this paper proposes a human blood pressure prediction model based on a real-valued deep belief network for predicting blood pressure changes in patients with hypertension and high-risk groups after a follow-up cycle.This model automatically extracted the influencing factors of future blood pressure through the deep network formed by the restricted GG-RBM stacking with double-layer Gaussian structure.It dug the complex and nonlinear relationship between blood pressure and its influencing factors.Based on adaptive distance estimation algorithm,the optimization process of parameter space was accelerated.The experimental results prove that for the prediction of patients’future blood pressure,the RDBN-based blood pressure prediction model has higher prediction accuracy than traditional prediction methods.