Research On Energy Expenditure And Prediction Model Of Men’s Badminton Players Based On Artificial Neural Network

Research purpose: The current research on triaxial accelerometers mainly involves low and medium intensity periodic sports,and less research on high intensity non-periodic sports and cannot accurately predict energy consumption,which is necessary to verify the effectiveness of triaxial accelerometers in measuring energy consumption of high intensity non-periodic sports.In this study,we measured the energy consumption of nonperiodic badminton sports and understood the basic characteristics of the sport,and used artificial neural network technology to build an energy consumption prediction model to improve the accuracy of energy consumption measurement so as to better guide athletes in physical training and technical and tactical training,and to enrich the measurement field of triaxial accelerometer so that it can better serve for sports practice.Research Methods: In this experiment,30 male badminton players were selected and divided into 24 in the training set and 6 in the validation set in a random cut according to the ratio of 8:2,and a portable cardiorespiratory fitness device combined with a triaxial accelerometer was used to monitor the energy consumption changes in a simulated game..The data were analyzed and processed using SPSS software to perform descriptive analysis of the energy metabolism substrate during the simulated game;the original accelerometer characteristic index data were combined with high-dimensional modeling techniques in the form of nonlinear mapping using Python programming language to construct the prediction model;the normal distribution test was used to determine the normality of the data,and independent sample t-test and one-way ANOVA were applied to The model predictions and calibration values were compared and analyzed;the generalization ability,consistency results and accuracy of the model were analyzed and compared by mean absolute error,root mean square error and scatter plot.Research results:(1)The average heart rate of the athletes in the badminton simulation game was 163beats/min,the intensity of activity was 11.26 METs,the energy expenditure was13.96kcal/min;the oxygen uptake was 2.88L/min,the carbon dioxide excretion was2.69L/min;the carbohydrate consumption was 9.89kcal/min,the fat consumption was2.99kcal/min The main energy-supplying substance of the subjects in the simulated race was carbohydrate,with carbohydrate consumption accounting for 70% of the total energy consumption,fat energy-supplying substance accounting for 21% and protein energysupplying substance accounting for 9%.(2)The predictions of energy expenditure and activity intensity by the built-in algorithm of the triaxial accelerometer were significantly different from the calibration values(P<0.01)and showed significant overestimation and underestimation.The results of the built-in algorithms for WW,F,and FC energy consumption were underestimated,while the results of the built-in algorithms for FVM3 and FVM3 C energy consumption equations were overestimated for the wrist,waist,and thigh,and underestimated for the upper arm and ankle;the results of the built-in algorithms for activity intensity were underestimated.(3)There was a significant difference between the predicted and calibrated values of the stepwise regression prediction equations based on SPSS for energy expenditure and activity intensity(P<0.01),with the mean absolute error of 0.1305-0.1798 for the energy expenditure equation and 0.1362-0.1704 for the activity intensity equation.(4)The MLP neural network was used to identify the relevant parameters flexibly by feature extraction,and the 49-17-1 three-layer network structure with the activation function of Tanh,the optimizer of Adam,the learning rate of 0.001 and the maximum number of iterations of 3000 was finally determined for the energy consumption and activity intensity models.(5)The mean absolute errors of the multiple linear regression model and the MLP neural network model were 0.0907 and 0.0502,respectively,in the prediction results of energy expenditure of athletes;the mean absolute errors of the multiple linear regression model and the MLP neural network model were 0.1070 and 0.0769,respectively,in the prediction results of activity intensity of athletes;by one-way ANOVA,Bland-Altman plots and error results analysis revealed that the MLP neural network model had better robustness,generalization ability and prediction accuracy.Research conclusions:1.the average heart rate of athletes in the simulated game was 163 beats/min,energy consumption was 13.96kcal/min,activity intensity was 11.26 METs,carbohydrate consumption accounted for 70% of total energy consumption,fat energy-supplying substances accounted for 21%,protein energy-supplying substances accounted for 9%;2.The three-axis accelerometer built-in supporting algorithm and the stepwise linear regression equation established through the acceleration of the original activity meter values for non-periodic sports such as badminton energy consumption and activity intensity can not be accurately monitored,it is necessary to add more indicators or even discover the more abundant and important information behind the original accelerometer to improve the prediction accuracy.3.A 49-17-1 three-layer network structure with an activation function of Tanh,an optimizer of Adam,a learning rate of 0.001,and a maximum number of iterations of 3000 is determined for the energy consumption and activity intensity model.4.The prediction model is constructed by combining the original acceleration characteristics index with energy consumption and activity intensity through Python programming language based on MLP neural network.Compared with the traditional model algorithm,the MLP neural network model has better robustness,generalization ability and prediction accuracy,and is more suitable for energy consumption monitoring of badminton sports.