| Objective:At present,most of our country’s research on energy consumption is centered around simple medium and low-intensity physical activity,and rarely involves high-intensity and complex physical activity,moreover,there are few applications of neural network model in predicting energy expenditure of physical activity.This study establishes different methods for processing accelerometer raw data of different wearing parts.The linear regression model and the neural network model of the wearing part are compared horizontally with the two models of the same and different wearing parts to find the optimal prediction model,thereby enriching the measurement field of the accelerometer,so that it is better Sports practice services.Methods:Subjects wore four Actigraph GT3X accelerometers(GT3X)on four sites(outside of the dominant and non-dominant wrists,the right hip and the right ankle),and one Cosmed K4b~2 gas metabolism energy was worn on the chest.Analyzer(referred to as K4b~2),through the large-screen playback of the clipped high-intensity aerobics video,remove the heat machine preparation and warm-up time,the subject carries a high-intensity complex exercise of 9 minutes(this study chooses fitness operation as typical representative).A general linear regression model was established using SPSS 22.0.The neural network model was established by SPSS Modeler 18.0,and the prediction accuracy of the two prediction models was tested using the Bland-Altman method and comparing the RMSE indicators.Results:(1)In the correlation analysis between body morphological indexes and energy consumption EE during exercise,the correlation between body weight index and EE was the highest,modeling group r=0.505(p<0.01),verification group r=0.41(p<0.01);In the correlation analysis between the accelerometer VM value of the four wearing parts and the EE and METs of the HR and K4b~2 energy consumption,the correlation between the VM values??of the four parts and the EE and METs was significant(p<0.01),among which the dominant wrist The correlation was the most significant.The correlation between HR and EE and METs was the strongest.The correlation with EE was 0.486(p<0.01),and the correlation with METs was0.502(p<0.01).(2)The linear regression prediction equations of the four parts constructed in this study are as follows:Non-dominant wristEE=0.000014vm+0.135 weight+0.004hr-1.071ankleEE=0.000025vm+0.135 weight+0.004hr-1.157waistEE=0.000029vm+0.135 weight+0.003hr-0.674Advantage wristEE=0.000024vm+0.133 weight+0.004hr+0.187After the adjustment of the non-dominant wrist,r~2=0.519,r~2=0.521 after the ankle adjustment,r~2=0.518 after the waist adjustment,and r~2=0.522 after the adjustment of the dominant wrist.(3)The initial learning rate of the three-layer neural network prediction model constructed in this study is 0.05,the momentum constant is set to 0.5,and the error rate is set to 0.001.The model is as follows:Non-dominant wrist 13-9-1 three-layer neural network model(r~2=0.808)Three-layer neural network model of ankle 13-9-1(r~2=0.796)Three-layer neural network model of the waist 13-13-1(r~2=0.771)Advantage wrist 13-8-1 three-layer neural network model(r~2=0.795)(4)In the consistency test between the measured value and the budget value of the linear regression model and the neural network model,in the ba diagram,the scatter points of different parts of different models basically fall between±1.96SD,non-dominant wrists,ankles,The neural network model and the linear regression model of the four parts of the waist and the dominant wrist have good predictive ability.(5)By calculating the RMSE,MAPE and BISA indices of different parts and comparing them horizontally,the RMSE and MAPE of the four parts of the non-dominant wrist,dominant wrist,ankle and waist are lower than the linear regression model.The network RMSE indices are:0.98,2.85,1.05,and 1.16,respectively.The RMSE indices of the linear regression model are:1.65,4.7,1.64,and 1.7.The dominant wrist and waist BISA indices of the neural network model are also significantly lower than the linear regression model.From the perspective of overall error,the error of the neural network model is smaller.Conclusions:(1)The linear regression energy consumption equations of the four parts constructed in this study have a high degree of fitness,and have high accuracy in predicting the energy consumption of motion.It can be applied to the monitoring of most high-intensity complex sports energy consumption.(2)The three-layer neural model constructed in this study has a high degree of fitness,and the accuracy is better than the general linear regression model.It can accurately predict the energy consumption of high-intensity complex motion,and it is verified that it predicts the kinetic energy.The consumption aspect is highly accurate and is the optimal prediction model. |
PDF Full Text Request