神经网络模型预测纵跳时下肢关节内力矩  

Prediction on Lower Limbs' Joint Torques Using Artificial Neural Network during Vertical Jumps

作　　者：刘宇[1] 史世民[1] 钟运健[1] 

机构地区：[1]上海体育学院,上海200438

出　　处：《体育科学》2008年第7期45-50,65,共7页China Sport Science

基　　金：国家自然科学基金(10672105);上海市科学技术委员会浦江人才计划项目(06PJ14082);上海市教育委员会重点理科课题项目(05ZZ42);上海市重点学科建设项目(T0901)

摘　　要：关节肌肉力矩在生物力学研究中是一个非常有价值且应用性颇广的参数。要获得关节力矩数值,需同步测量运动学和地面反作用力参数,并通过逆向动力学(Inverse Dynam- ic)过程方能求得。目的在于建立推估两种纵跳:直立(Counter-Movement Jump,CMJ)和蹲位(Squat Jump,SJ)时下肢各关节力矩的人工神经网络(artificial neural network,ANN),通过输入相对较容易测得的地面反作用力参数,推估下肢关节力矩(踝、膝、髋)。以10位男性运动员为受试者,在测力台上完成两种纵跳,其所测得的地面反作用力参数作为神经网络模型的输入数据,以下肢各关节力矩为输出数据。经过网络优化后,所求得的最佳神经网络模型为"5-10-3模型"(即输入层有5个神经元;隐藏层为1层,有10个神经元;输出层有3个神经元)。该模型所推估出的下肢各关节力矩极值相对误差均小于6%,与实测值的相关系数高达0.95以上,且在推估关节力矩曲线形态与角冲量方面的表现也十分良好,显示通过神经网络以地面反作用力来推估下肢各关节力矩的方法是准确和可行的。The joint torque is one of the meaningful and useful biomechanical parameters in biomechanics. However, in order to obtain the joint torque and inverse dynamics by inputting kine matic, the ground reaction force （GRF） and anthropometric data must be conducted. The purpose of this study was to develop an artificial neural network （ANN） for predicting the joint torques of lower limb using solely the GRF parameters for counter-movement jump （CMJ） and squat jump （SJ）. Ten male sport students performed CMJ and SJ on force platform, meanwhile the kinematics data were recorded and the joint torques was calculated as experimental data by inverse dynamics. We used a fully-connected, feed-forward network comprised of one input layer, one hidden layer and one output layer trained by back propagation using Steepest Descent Method. The input parameters of ANN were relevant time variables of GRF and the output parameters were joint torques of lower limb. After trial-and-error optimization procedure, a ＂5- 10-3 ANN model＂ was developed. The results revealed that, compare to the measured joint torque, the relative errors of predicted peak values were less than 6 % and the correlation coefficients between them were greater than 0.95 for all joints. The curve patterns and the angular impulses of the predicted torques were also very similar with the measured torques, indicating that the ANN model fitted the experimental data well and the model is feasible in assessment of joint torque for CMJ and SJ.

关 键 词：纵跳 关节力矩 人工神经网络 逆向动力学 

分 类 号：G804.6[文化科学—运动人体科学]