Signal Tracking Of Nonlinear System Based On Dimension Compressed Neural Network

In this paper,we use a neural network with dimension compression to track the nonlinear system.Because the nonlinear systems widely used in many industries have complex uncertainties and internal and external interference terms,the precise model of the system is generally very rare.Therefore,compared with traditional control,neural network control has become an excellent alternative because of its strong fitting ability.The controller based on neural network can be used to compensate some time-varying nonlinear parts and uncertainties in the system.But it is well known that the fitting ability of the neural network is guaranteed by enough training samples,but for the online control process,the requirement has evolved into enough center points.We also know that the number of nodes is related to the dimension of a nonlinear system.So this paper first discusses the signal tracking application of erebellar model articulation controller(CMAC)neural network in Chua's circuit.This is a kind of lightweight network,and the few nodes bring the lack of fitting ability,and based on the principle of robot control design,it is difficult to apply to the control of other nonlinear systems.Then we discuss the control of the Radial-Basis Function(RBF)neural network on the nonlinear system of the robotic arm,using the adaptive control and using the backstepping method.After comparing with the CMAC neural network,we can see that the number of nodes in the RBF neural network is enough,and as a widely used neural network,its fitting ability is stronger and the control effect of the nonlinear system is better.But too many knots will bring huge system burden.Especially when the number of nodes increases exponentially with the increase of system dimension,it will seriously restrict the development of neural network control technology.In order to solve the defect of the increase of node exponent in neural network,the structure of neural network is improved by the idea of dimension compression,and it is verified in mathematics.Finally,the validity of the theory is verified by simulation,and compared with the CMAC neural network and the RBF neural network,it shows that the dimension compression neural network has better fitting ability and very light structure.Finally,in order to verify the comparison with CMAC,the dimension compression RBF neural network designed in this paper also has certain applicability in other classical nonlinear systems.This paper also applies it to the speed control of induction motor to achieve satisfactory signal tracking effect.