肌肉—关节模型在人体动态平衡建模中的应用

发布时间：2018-03-05 05:34

本文选题：人体动态平衡　切入点：肌肉-关节模型　出处：《河北大学》2017年硕士论文　论文类型：学位论文

【摘要】：平衡能力是人类进行各项活动的基础,包括静态和动态平衡能力,其中人体平衡的研究趋势为动态平衡。人体动态平衡依赖于前庭觉、视觉、本体感觉以及中枢神经系统等多方面调节与控制来完成。由此可见,其调节过程是一个较为复杂的非线性系统。而建立人体动态平衡的非线性模型对于深入研究人体动态平衡和提高康复医疗水平具有重要的推动作用。但是国内外的研究大多针对人体运动系统进行建模,而在动态平衡建模方面的研究甚少。本文将肌肉-关节模型应用到人体动态平衡的建模中,解决了人体动态平衡的非线性建模问题,研究内容包括以下几个方面:第一,利用无线肌电采集系统,采集人体在静止和随机视觉激励状态下的下肢肌电信号,经过预处理并提取时域特征参数-均方根值和积分肌电值,对两种状态下的下肢肌肉活跃程度进行了对比分析。结果表明:股直肌和胫骨前肌在两种状态下的均方根值无显著差异;股直肌与胫骨前肌在静止站立实验中肌肉的贡献率很低,且在随机视觉激励状态下其贡献率表现为明显下降趋势。第二,根据人体动态平衡的调节机理,确定人体动态平衡调节系统的输入输出,在随机视觉激励下采集人体动态平衡建模所需数据,在建模过程中引入BP神经网络的非线性辨识方法,建立肌肉-关节模型,实现对人体动态平衡调节系统的建模;而由于模型精度的不理想,依据下肢肌肉均方根值与踝关节角度的关联性,依次去除训练集中与踝关节角度相关性较低肌肉的均方根值,重新采用非线性辨识方法对人体动态平衡的调节系统进行建模,找出更为理想的辨识方案。第三,由于SVM较强的泛化能力,能有效地避免BP神经网络陷入局部极小的缺点,本研究依据优化后的辨识方案,采用SVM的非线性辨识方法,对人体动态平衡的调节系统进行建模,同时与BP神经网络优化后的预测结果进行对比。结果表明:SVM预测输出与期望输出之间的相关系数要高于BP神经网络预测与输出的相关系数,同时SVM的预测误差比BP神经网络预测误差要小。因此,在人体动态平衡的预测方面,SVM预测效果总体上优于BP神经网络。总之,本研究实现了人体动态平衡的非线性建模,并在一定程度上提高了人体动态平衡调节过程的非线性辨识精度,为医学研究提供了理论依据与数据支撑。
[Abstract]:The ability to balance is the basis for all human activities, including static and dynamic balance, in which the research trend of human balance is dynamic balance, which depends on vestibule, vision, Proprioceptive sensation and central nervous system are regulated and controlled in many ways. The regulation process is a complex nonlinear system, and the establishment of a nonlinear model of human dynamic balance plays an important role in studying the dynamic balance of human body and improving the level of rehabilitation medical treatment at home and abroad. Most of the research in this paper focuses on the modeling of human motion system, In this paper, the muscle-joint model is applied to the modeling of human dynamic balance, and the nonlinear modeling problem of human dynamic balance is solved. The research content includes the following aspects: first, Using wireless electromyography acquisition system, the lower limb electromyography (EMG) signals of human body under static and random visual excitation were collected, and the time domain characteristic parameters, root mean square (RMS) and integral electromyography (EMG), were extracted after pretreatment. The results showed that there was no significant difference in RMS between the rectus femoris muscle and the anterior tibial muscle. The muscle contribution rate of rectus femoris and anterior tibial muscle in static standing test is very low, and its contribution rate is obviously decreasing under random visual excitation. Secondly, according to the regulation mechanism of human dynamic balance, The input and output of the human dynamic balance regulation system are determined, and the data needed for the modeling of human dynamic balance are collected under random visual excitation. The nonlinear identification method of BP neural network is introduced in the process of modeling, and the muscle-joint model is established. To realize the modeling of the human dynamic balance regulation system, and because the model accuracy is not ideal, according to the correlation between the lower extremity muscle RMS value and the ankle angle, the RMS value of the training concentration and the ankle joint angle is removed in turn, and the RMS value of the lower muscle with the lower ankle angle correlation is removed in turn. The nonlinear identification method is used again to model the regulating system of dynamic balance of human body, and a more ideal identification scheme is found. Thirdly, because of the strong generalization ability of SVM, the shortcoming of BP neural network falling into local minima can be effectively avoided. According to the optimized identification scheme and the nonlinear identification method of SVM, the model of human dynamic balance regulating system is established in this paper. The result shows that the correlation coefficient between the predicted output and the expected output is higher than the correlation coefficient between the prediction and the output of BP neural network. At the same time, the prediction error of SVM is smaller than that of BP neural network. Therefore, the prediction effect of SVM is better than that of BP neural network. To a certain extent, the nonlinear identification accuracy of human dynamic balance regulation process is improved, which provides theoretical basis and data support for medical research.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】