基于自适应模糊神经网络的下肢关节角度估计  

Lower Limb Joint Angle EstimationBased on Adaptive Fuzzy Neural Network

作　　者：刘克平[1] 滕召纬 孙中波 李婉婷 LIU Ke-ping;TENG Zhao-wei;SUN Zhong-bo;LI Wan-ting(Changchun University of Technology,School of Electrical and Electronic Engineering,Changchun Jilin 130012,China)

机构地区：[1]长春工业大学电气与电子工程学院,吉林长春130012

出　　处：《计算机仿真》2022年第9期456-461,共6页Computer Simulation

基　　金：国家自然科学基金(61873304,11701209);中国博士后科学基金:(2018M641784,2019T120240);吉林省教育厅科学研究(JJKH20210745KJ)。

摘　　要：针对人体下肢关节角度预测精度不足问题,提出一种结合神经网络和模糊推理理论的下肢关节角度预测方法——自适应模糊神经网络(Adaptive Fuzzy Neural Network)。模型的输入数据为采集和处理后的人体下肢股直肌(Vastus Rectus)、股外侧肌(Vastus Lateralis)和长伸肌(Extensor Pollicis Longus)的表面肌电信号,输出结果为预测得到的髋、膝、踝三个关节角度。同时在算法迭代更新过程中,结合混合策略和误差反向传播算法完成对结构参数的自适应调整与优化,进而提高模糊神经网络模型的预测精度。仿真结果表明,所提AFNN方法相比BP和RBF神经网络预测模型,能够实现较好的人体下肢三关节角度的预测,验证AFNN方法可以有效提高下肢三关节角度预测精度。In order to solve the problem of insufficient accuracy of human lower limb joint angle prediction, an adaptive fuzzy neural network(AFNN) combining neural network and fuzzy reasoning technology is proposed in this paper. The input data of this model are the sEMG signals of vastus rectus muscle(VR),vastus lateralis muscle(VL) and extensor pollicis longus(EP) of human lower limbs after collection and processing, and the output results are the predicted joint angles of hip, knee and ankle. During the iterative update process of the algorithm, the hybrid strategy and error back propagation algorithm are adopted to realize the adaptive adjustment of structural parameters, which improves the prediction accuracy of the model. The simulation results show that the proposed AFNN method is better than BP and RBF neural network in predicting the angle of the lower limb joints, it is verified that the AFNN method proposed in this paper can effectively improve the prediction accuracy of lower limb three joint angles.

关 键 词：运动意图 下肢关节角度估计 表面肌电信号 自适应模糊神经网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]