基于改进PNN模型的人体健康监测方法

Monitoring Method for Human Health Based on Improved PNN Model

远程人体健康监测分析呈现滞后性、不准确性、设备昂贵等特点,因此难以实现实时、准确的人体健康监测分析。通过对人体健康监测方法和概率神经网络(PNN)的研究,将具有调节参数少、收敛速度快和保证获得贝叶斯最优解等优点的PNN应用于人体健康监测。但是PNN的缺点是未考虑不同类别模型之间的重叠和交错,以及当训练样本不满足假定条件时无法确定是否存在相应的PNN模型。针对这两个缺点,分析了径向基神经网络(RBNN)和广义回归神经网络(GRNN)的网络拓扑结构和优势,创新性地提出在PNN结构的模式层中引入RBNN结构,以及在PNN结构的输出层中引入GRNN结构,得到了一种新的径向基-广义回归-概率混合神经网络(RBF-GR-PMNN),从而满足实时、准确监测人体健康状况的要求。进行了RBF-GR-PMNN与一般PNN的对比试验。试验分别从准确率和运行时间等方面进行对比分析。试验结果证明了改进PNN在这些方面均优于一般PNN,进一步表明了RBF-GR-PMNN模型的有效性。

Remote monitoring and analysis of human health is characterized by lag,inaccuracy,and expensive equipment.Therefore,it is difficult to realize real-tim e and accurate human health monitoring and analysis.Through the research on the human health monitoring method and probabilistic neural network(PNN),the PNN is applied to human health monitoring with the advantages of fewer regulating parameters,faster convergence speed and guaranteed Bayesian optimal solution.However,the defects of PNN is that the overlapping and interleaving between different categories of models are not considered,and when the training samples do not satisfy the assumed conditions,it is impossible to determine whether there is a corresponding PNN model or not.Aiming at these disadvantages,the topologic structures and advantages of radial basis neural network(RBNN)and generalized regression neural network(GRNN)are analyzed;and the creative proposal is put forward,which is introducing RBNN into the model layer of the PNN structure and introducing the GRNN structure into the output layer of the PNN structure,thus a new radial basis function-generalized regression-probabilistic mixed neural network(RBF-GR-PMNN)is proposed to meet the requirement of real-time and accurate monitoring of human health.A comparative experiment between RBF-GR-PM NN and general PNN is carried out.The accuracy and running time of the experiments are compared and analyzed.The results show that the RBF-GR-PMNN is superior to the general PNN in various aspects.So the effectiveness of the RBF-GR-PMNN model is further demonstrated.