基于BP神经网络的18~30岁人群功率自行车能量消耗模型构建

Energy Consumption Model of Bicycle Ergometer for 18—30 Year Olds Based on BP Neural Network

目的:以心率为自变量根据不同算法建立适合的18~30岁普通健康成年人功率自行车能量消耗预测模型。方法:以39名18~30岁普通健康成年人为研究对象,分别完成3种强度(30 W、75 W、125 W)的功率自行车测试。受试者佩戴气体代谢分析仪和心率带,采集此过程中心率、摄氧量、二氧化碳呼出量等指标,并以此作为数据来源,进行相关性分析获得具有统计学意义的指标,分别建立能量消耗的线性回归模型和BP神经网络模型。采用Bland-Altman散点图对模型预测的准确性进行分析;采用均方误差(MSE)、平均绝对误差(MAE)、决定系数(R^(2))与平均绝对百分比误差(MAPE)评价模型的拟合和预测效果。结果:(1)能量消耗与%HR(实际心率/安静心率×100%)、体重、体脂百分比、身高存在相关性(P<0.01)。(2)建立的BP神经网络模型(包含4个输入层、3个隐含层和1个输出层)的最佳性能表现MSE为0.48,训练集、验证集、测试集和整体的R值分别为0.92、0.93、0.92和0.93。(3)神经网络模型与多元线性回归模型预测值相比差异无显著性(P>0.05),但神经网络预测模型的R^(2)高于现行回归方程,MSE、MAPE、MAE均低于线性回归方程。结论:采用BP神经网络可以提高预测精度,降低采用线性回归方程预测时的误差。

AIM:This research aims to establish a suitable prediction model for power cycling energy expenditure for ordinary healthy adults aged 18—30 years old based on different algorithms using heart rate as the independent variable.METHODS:Thirty-nine healthy adults were selected as study subjects and completed power cycling tests at three intensities(30 W,75 W,125 W).Subjects wore gas metabolic analyzers and heart rate bands to collect the indexes of heart rate,oxygen uptake,and carbon dioxide exhalation during this process.These data were used for correlation analysis to obtain statistically significant indexes and to establish linear regression models and BP neural network models of energy expenditure,respectively.The accuracy of model prediction was analyzed by using Bland-Altman scatter plot;mean square error(MSE),mean absolute error(MAE),coefficient of determination(R~2)and mean absolute percentage error(MAPE)were used to evaluate the fitting and prediction effects of the models.RESULTS:(1)there was a correlation between energy expenditure and%HR(actual heart rate∕resting heart rate×100%),body weight,body fat percentage,and height(P<0.01).(2)The MSE of the established BP neural network model(containing four input layers,three hidden layers,and one output layer)was 0.48,and the r-values for the training set,validation set,test set and ensemble were 0.92,0.93,0.92 and 0.93 respectively.(3)There was no significant difference(P>0.05)between the predicted values for the neural network model and multiple linear regression model,but the R~2 of neural network prediction model was higher than the current regression equation,and MSE,MAPE and MAE were lower than the linear regression equation.CONCLUSION:The use of BP neural networks can improve the prediction accuracy and reduce the error when using linear regression equations for prediction.