【作者】 李韬；

【导师】 王江；

【作者基本信息】 天津大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 近年来，国内外每年都有大量的关于 sEMG 信号分解的文章发表。他们研究的目标是将 sEMG 信号分解为 MUAP 或 MUAPT，通过动作模式识别控制假肢。很少有学者将 sEMG 信号分解为 SFAP。sEMG 信号分解至 SFAP 在疾病诊断、针灸治病机理分析以及假肢控制等方面具有重要意义。 本文在 Graupe D.和 Huang Q.等研究的基础上，提出了一种新的基于 RBF 神经网络和遗传算法的 sEMG 至 SFAP 的分解算法。Graupe D.等巧妙地运用了sEMG 信号可以近似用若干 Gaussian 函数的和表示的特点，以 Gaussian 函数作为 Hopfield 神经网络的节点函数，运用 Hopfield 神经网络拟合 sEMG 信号。通过曲线拟合，将 sEMG 信号分解为若干 Gaussian 函数，聚类处理后就得到了组成 sEMG 信号的 SFAP。本文基于同样的思路，采用具有全局逼近性能和最佳逼近性能的 RBF 神经网络拟合 sEMG 信号，避免了 Hopfield 神经网络易陷入局部最优解的缺点，加快了算法的收敛速度，提高了拟合的精度。为了提高算法的精度，本文采用参数优化能力强的遗传算法训练 RBF 神经网络的权值和节点函数参数。 另外，本文在肌肉电生理学研究的基础上提出了 sEMG 至 SFAP 的可分解性定理，指出了在轻度收缩条件下 sEMG 至 SFAP 可分解且分解结果唯一，为算法的设计提供了理论依据。更多还原

【Abstract】 Recently, many papers about sEMG signals decomposition were published athome and overseas. Their research aimed at decomposing sEMG signals into MUAPor MUAPT, and controlling artificial limb by action pattern identifying. Few peopledecompose sEMG into SFAP. The decomposition had great significance in diseasesdiagnosis, acupuncture analysis, artificial limb control, and etc. According the research of Graupe D and Huang Q, a new decompositionalgorithm of from sEMG signals to SFAP, which applied RBF neural networks andgenetic algorithm, was proposed in this paper. Graupe D and Huang Q ably appliedthe character that sEMG signals could be approximatively expressed by the sum ofseveral Gaussian functions. They approached sEMG signal by Hopfield neuralnetworks, whose node functions were Gaussian functions. During the process ofcurrent fitting, sEMG signals were decomposed into Gaussian functions, and theconstituent SFAPs were got by clustering these Gaussian functions. According thesimilar pathway, RBF neural networks, which had the characteristics of universalapproximation and optimal approximation, were applied to approach sEMG signals.The application of RBF neural networks avoided the drawback of Hopfield neuralnetworks’ easily getting struck at a local optimum. It also increased the decompositionconvergent speed, and improved the approaching accuracy. In order to improve theaccuracy of algorithm, genetic algorithm, with strong ability of parameters optimizing,was applied in this paper to train all the parameters of RBF neural networks. In addition, a theorem of the decomposability of sEMG signals into SFAP wasproposed in this paper on the basis of muscle electrophysiological study. It waspointed out in this paper that sEMG was decomposable and the result was exclusiveon the condition of light shrinkage, which provided a theories basis for decompositionalgorithm designing.更多还原