Abstract: Considering the lacking function of existing upper-limb rehabilitation training robot to automatically recommend a suitable training mode for the impaired limb, a prescription diagnosis method combining the support vector machine and generalized dynamic fuzzy neural networks (SVM-GDFNN) is proposed. Based on the principle to minimize structure risk, SVM does very well in generalization ability, which is used to recommend a preliminary prescription diagnosis for the sample. At the same time for the SVM method is prone to make wrong diagnosis in the support vector area, GDFNN is employed to conduct referral for the samples in the area, and then the training mode of impaired limb is prescribed according to the designed principles. Combining the clinical application, the feature extraction method for impaired-limb movement performance is analyzed, and the process of the SVM-GDFNN model building and sample diagnosis are presented in detail. Clinical experiment results indicate that the suggested method can effectively reduce the fault diagnosis and serve with a high diagnostic accuracy rate. By designing function to automatically recommend training mode, it is helpful to improve the clinical intelligent level of rehabilitation training robots.

Key words: Generalized dynamic fuzzy neural networks, Intelligent diagnosis, Rehabilitation robot, Support vector machine