Repetitive Neural Networks With Its Relevant Applications

Recurrent neural networks are a kind of parallel computing method,and are able to process data in a high speed,that makes it used to solve the problem of time-varying matrix in real time.The existing recurrent neural networks work in time domain infinitil.While,the periodic time-varying problem is significant in finite interval of time.Periodic time-varying matrix inversion and periodic quadratic programming problem are a kind of universal periodic time-varying matrix.Periodic time-varying matrix inversion includes periodic time-varying phalanx,periodic time-varying matrix generalized inverse solution.Redundant manipulator trajectory planning problem is a kind of periodic time-varying problems in application.To solve the periodic time-varying problem,repetitive neural networks are proposed in this thesis.Comparedf with the recurrent neural networks,the repetitive neural networks are appropriate for solving periodic time-varying problems.The main work of this thesis is summarized as follows:1.Based on the recurrent neural networks,a repetitive neural network is put forward.The convergence is analised by Barbalat-Like lemma.2.To solve the periodic time-varying square matrix inverse problem,a repetitive neural network is suggested.Furthermore,the corresponding repetitive neural network solving model is gotten.According to the model above,a simulation is given with a practical example.3.In view of the periodic time-varying matrix generalized inverse problems,a repetitive neural network is suggested.Furthermore,the corresponding repetitive neural network solving models are gotten for the left pseudo-inverse and the right pseudo-inverse.Considerating pseudo inverse examples respectively,Matlab/Simulink models which are depended on the solving models above are constructed.Meanwhile,linear activation function and hyperbolic activation function are used in the models.4.In view of the periodic time-varying quadratic programming problems,a repetitive neural network is suggested.Furthermore,the corresponding repetitive neural network solving model is gotten.According to the model above,a simulation is given with a practical example.5.The repetitive path planning is reformulated as a quadratic programming problem.Based on the periodic operation task of redundant manipulators,the repetitive neural networks are used to solve it.The example of five-linked planar manipulators is simulated to prove the effectiveness of repetitive neural networks.