The Application Of Neural Network Method In Predicting The Energy Consumption Of Physical Activity

Objective:there are many ways to measure physical activity in the body at present,but there are considerable limitations.The energy consumption of physical activity measured by portable acceleration sensor has developed into the mainstream,but the algorithm model for predicting energy consumption of physical activity is generally linear regression model,and its prediction ability is limited.In this study,we hope to obtain the acceleration data by analyzing the three axis acceleration sensor of the hip,and establish an artificial neural network(Artificial Neural Network,ANN)model to predict the energy consumption.The converted signal information and the population characteristics of the subjects(sex,age,height,body weight,body mass index)are applied to the training of the artificial neural network(ANN)model,trying to find a better nonlinear mathematical model to analyze the complex relationship between variables,thus reaching the effect of predicting energy consumption.Methods:according to the search of the literature,the subjects carried out the daily life movements such as 65m/min,100m/min,135m/min,sitting,and bicycle riding(middle intensity heart rate),and recorded the subjects,age,sex,height,weight,fat content,acceleration sensor count(VM).For the main indicators,the subjects were asked to fill in the basic situation questionnaire for age and sex,and the height and weight were measured according to the measurement standard of the body morphological index;the fat content was measured by the body composition analyzer.The three axis accelerometer(GT3X)reads raw count(VM)to collect one minute data for acceleration signal feature extraction.A GT3X acceleration sensor is fixed on the hip of the subjects,and 5 kinds of activities are started.The sampling frequency of the accelerometer is set to 30Hz,and the signal is collected by the GT3X software.Matlab2015b is used to read the original acceleration signal for correlation analysis.The energy consumption was measured by indirect calorimetry(MAX-II),and the subjects collected gas samples by wearing breathing masks.The purpose of establishing ANN model is to combine neural network with original data characteristics to evaluate energy consumption per minute.Results:the model prediction results show that the MAE and MSE mean values of the ANN model are significantly reduced compared with the regression equation of Actigraph(P<0.001)the correlation coefficient(R~2)of the artificial neural network(ANN)is higher than the AG regression equation model,while the average absolute error(MAE),the mean square error(MSE),and the total energy consumption(TEE)are higher than that of the AG regression equation model.The absolute percentage difference is less than that of AG regression equation model.The mean and variance of the predicted values have been improved.The percentage error analysis shows that the ANN model has a further reduction than the AG regression equation model.In this paper,the original signal data of acceleration sensors are collected on the basis of the existing basis,and the data are used in the feature extraction stage to establish the relevant parameters for modeling.After analysis,the following conclusions are obtained.(1)the ANN energy consumption prediction model established in this study improves the prediction accuracy of energy consumption and is more suitable for monitoring the daily energy consumption of college students.(2)opening up the field of high-dimensional modeling technology,compared with the traditional modeling technology,ANN is more flexible.It is verified that the current model also makes the model algorithm more robust.