基于PSO_GRNN网络的肺内静态压力值预测方法

The method of pulmonary static pressure value prediction based on PSO_GRNN network

为了实现机械通气辅助医疗中依据病人个体差异控制呼吸机通气参量,分析了基于广义回归神经网络(GRNN)的呼吸系统力学模型,通过结合PSO_GRNN网络、数值积分和递推最小二乘法等实现呼吸系统模型的参数辨识。采用直接计算法实现单周期呼吸样本的肺静态压力值计算,并利用二阶多项式拟合体积值误差,计算10个吸气周期静态数据点的平均绝对值误差为0.1693 mL,计算10个呼气周期静态数据点的平均绝对值误差为0.3728 mL。采用PSO_GRNN网络实现多周期呼吸样本集的肺静态压力值预测,10个呼吸周期样本集的训练集平均误差为0.0009 kPa,测试集平均误差为0.0407 kPa。仿真实验结果表明PSO_GRNN网络在收敛速度、平均误差、运算速度等方面均优于PSO_BP网络。所用方法在机械通气辅助治疗时可以为医生设置呼吸机通气参量提供有效的参考依据。

It needs to control ventilator parameters according to individual differences of patient in the auxiliary treatment of mechanical ventilation.In this study,the mechanical model of a respiration system based on general regression neural network(GRNN)are analyzed.To identify parameters of the respiratory system model,a fusion method based on PSO_GRNN,numerical integration and recursive least square is proposed.The static lung pressure value of single-cycle respiratory samples is calculated by direct calculation and the second order polynomial is used to fit the volume error.The mean absolute error of static data points for ten inhalation cycles is 0.1693 mL,and the mean absolute error of static data points for ten expiratory cycles is 0.3728 mL.PSO_GRNN is used to predict the static lung pressure of the multi-cycle respiratory sample set.For the ten sample sets of respiratory cycle,the average error of the training set is 0.0091 and the average error of the test set is 0.4065.Simulation results show that PSO_GRNN is better than PSO_BP in terms of convergence rate,average error and computation speed.The proposed method can provide an effective reference basis for doctors to set ventilator parameters during the mechanical ventilation treatment.