Construction And Evaluation Of Physical Activity Energy Consumption Segmentation Model Based On Accelerometer Data And Heart Rate Signal

Objective:At present,most of the energy consumption prediction models constructed in China through accelerometers or heart rate data are mainly for running activities.The accuracy of predicting two or more types of physical activity is insufficient.Accelerometer data and heart rate signals have their own advantages.Disadvantages.Few studies have combined the two to predict different levels of physical activity.This study collects VM values and HRnet data of accelerometers worn at different parts,constructs segmented models of different parts,and evaluates them with other models.Exploring the best wearing parts and the most ideal segmented models,and then enriching the model prediction field of accelerometer and heart rate data,so that it can better serve daily sports practice.Methods:(1) The study adopts a measurement method,taking 50 non-sports students as subjects,and dividing them into a modeling group (n=30) and a verification group (n=20).Wear Actigraph GT3X accelerometer in 3 places (right wrist,chest and hip),wear Cosmed K4b2 gas metabolism analyzer and Polar H10 heart rate chest strap to connect K4b2 to monitor heart rate data;modeling group and verification group were completed separately:use 15 types of physical activities,including standing,standing,sitting upright,computer typing,etc.(2) Using the data of IC as the calibration value,using the VM value and HRnet data of different parts to determine the threshold,using stepwise regression to build VM and HRnet models of different parts,and using the validation group data to model R2,RMSE,r2 Indicators for reliability verification;Results:(1) In this study,VM values and HRnet were combined into LC values to predict mild and moderate physical activity,and HRnet predicted severe physical activity.The threshold cut-off points for HRnet and LC values were:HRnet=9,wrist LC=0.14,LC=0.11 in the chest,LC=0.12 in the hip;three segmental models constructed were:Segmented model wrist EE=(?)Segmented model chest EE=(?)Segmented model hip EE=(?)The constructed single linear models R2 of VM and HRnet are:LinearVM wrist model (R2=0.575);LinearVM chest model (R2=0.706);LinearVM hip model (R2=0.676);LinearHRnet model (R2=0.501);GT3X Sasaki wrist model (R2=0.693);GT3X Sasaki chest model (R2=0.720);GT3X Sasaki hip model (R2=0.693);(2) After verification,the results show that the three VM-HRnet segmented models can accurately predict light and medium heavy physical activities without large error offsets.On the whole,the chest is the best wearing position for all models;although the LinearHRnet model performs stable in the prediction and does not show a large error offset,it generally underestimates the EE;The LinearVM model and the GT3X Sasaki model are generally mild Overestimate EE during moderate physical activity and underestimate EE during heavy physical activity.Conclusion:(1) This study uses the accelerometer VM value combined with the heart rate indicator HRnet to synthesize the LC value,divide the LC value of different parts and the threshold cut-off point of HRnet,and establish a segmented energy consumption prediction model.Light,medium,and heavy physical activity,and the segmentation model has good prediction capabilities;(2) Linear VM model and GT3X Sasaki model will be overestimated when predicting mild to moderate physical activity,the biggest error is predicted for riding a power bicycle,and Linear HRnet model is generally underestimated EE,which indicates a single linear model There will be certain defects in prediction,and in daily use,a suitable model should be selected according to the type of project.