基于人工神经网络和摄氧量动力学的身体活动能量消耗预测

Predictive Modeling of Physical Activity Energy Expenditure using Artificial Neural Networks and Oxygen Uptake Dynamics

目的:基于人工神经网络和摄氧量动力学,以心率(heart rate,HR)和呼吸频率(breathing frequency,BF)为自变量建立准确的身体活动能量消耗(physical activity energy expenditure,PAEE)预测模型。方法:招募24名健康大学生(年龄22.2岁±2.0岁,身高174 cm±9 cm,体重67.1 kg±12.4 kg),参加3次运动测试,包括递增负荷运动、恒定负荷运动(40%和70%VO_(2)max)和4种身体活动(6.4 km/h步行、体感游戏、负重行走、上下楼梯)。采集运动中HR、BF和摄氧量(VO_(2)),并建立PAEE预测的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,并计算了平均绝对误差(mean absolute error,MAE)和平均绝对百分比误差(mean absolute percentage error,MAPE)以评估模型预测的误差。结果:建立的模型在步行活动中,模型预测的VO_(2)(18.50±1.19 mL/kg/min)与观测值(18.34±1.44 mL/kg/min)接近,MAE=1.31 mL/kg/min,MAPE仅为7.32%。在体感游戏中,模型预测(19.31±1.22 mL/kg/min)与观测值(17.81±2.44 mL/kg/min)存在较大差异,其MAPE仍然控制在13.04%。对于负重行走和上下楼梯,MAPE分别为14.21%和11.12%。Bland-Altman分析结果显示系统偏差为0.1068 mL/kg/min,表示预测VO_(2)略高于标准VO_(2)。结论:基于摄氧量动力学构建的BP神经网络模型可以较为准确地预测PAEE,展示了HR联合BF在预测PAEE方面的潜力。

Objective:This study aimed to devise an accurate predictive model for Physical Activity Energy Expenditure(PAEE)by leveraging artificial neural networks in conjunction with oxygen uptake kinetics,utilizing heart rate(HR)and breathing frequency(BF)as predictors.Methods:We engaged 24 healthy university students(mean age:22.2±2.0 years,height:174±9 cm,weight:67.1±12.4 kg)in a comprehensive exercise protocol.This encom⁃passed three distinct exercise tests:graded exercise tests,constant load exercises(40%and 70%VO_(2)max),and four varied physical activities(6.4 km/h walking,exergaming,loaded walking,and stair ascent/descent).Throughout these activities,we continuously recorded HR,BF,and oxygen uptake(VO_(2))to subsequently establish a BP neural network model for PAEE prediction.The prediction errors of model’s performance was calculated using the Bland-Altman scatter plots,mean absolute error(MAE),and mean absolute percentage error(MAPE).Results:For the 6.4 km/h walking,the model’s VO_(2) prediction(18.50±1.19 mL/kg/min)is close to the observed VO_(2)(18.34±1.44 mL/kg/min),yielding an MAE of 1.31 mL/kg/min and the MAPE of 7.32%.During the exergaming session,the model prediction(19.31±1.22 mL/kg/min)exhibited a noticeable deviation from the observed value(17.81±2.44 mL/kg/min),though the MAPE remained at 13.04%.During loaded walking and up/down stair activities the MAPEs were 14.21%and 11.12%respectively.Bland-Altman analysis showed a systematic deviation of 0.1068 mL/kg/min,which indicated that the predicted VO_(2) was slightly higher than the standard VO_(2).Conclusion:The BP neural net⁃work model based on oxygen uptake dynamics can accurately predict PAEE,which shows the potential of HR com⁃bined with BF in the prediction of PAEE.