Font Size: a A A

Study On Control Of Nonlinear Uncertain System By Neural Network

Posted on:2015-01-24Degree:DoctorType:Dissertation
Country:ChinaCandidate:H Q RenFull Text:PDF
GTID:1268330428984065Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Neural networks have been widely applied in the intelligent control domain ofnonlinear uncertain systems for its strong learning ability and nearly universalapproximation ability. This thesis mainly research the intelligent control methods ofnonlinear uncertain systems based on neural network, analyze the principle and thesimulated control of nonlinear uncertain systems by traditional neural network,improved neural network structure, improved learning algorithm and updatingalgorithm of control, then the control processes are simulated. The main contents ofthis thesis are as follows:1. Receding horizon optimal control is an effective nonlinear adaptive controlmethod, firstly, the Neural Generalized Predictive Control(NGPC) based on neuralnetwork is researched in this thesis, On the basis of analyzing of the advantages anddisadvantages and algorithm efficient, the neural inverse control and the traditionalNGPC are combined to propose a kind of receding horizon optimal control schemewhich has the ability of prediction based on neural network, the novel logic structureof the control scheme is introduced. A “single pole double throw” is used in the newcontrol structure, which is used to realize the swift soft shift between the neural inversecontrol and the NGPC control scheme, and the improved updating control algorithm isdiscussed in detail. This scheme makes a better progress in control speed and precision.Then the improved scheme is used to control the Continuous Stirred Tank Reactor(CSTR)in simulation. The experiments results of the traditional control scheme andthe improved control scheme are illustrated together for comparison, and the resultsshow that the improved scheme is prior in performance to the traditional scheme fornonlinear uncertain systems control. 2. The principle of time delay neural networks with time delay neuron structure isdiscussed together with the spatial and temporal characteristics in this thesis, and theweight update algorithm of time delay neural network is given. Then the time delayneural network is embedded into the receding horizon optimal control schemementioned above. By inducting the time delay factor, the second order iterativederivative of the output error of different domain with respect to the input can becalculated in one cycle in the receding horizon optimization control scheme, which hasto be computed by iterations in too more cycles. The improved scheme shorten theduration of the receding horizon optimal control, improve the real time control of thespeed of receding horizon control scheme, improve the real-time control ability.Through the simulation experiment, the actual control performance of the improvedoptimized control scheme is demonstrated finally.3. The traditional back propagation algorithm (BP) of neural network is analyzedfrom the perspective of statistical mechanics in the thesis, a new adaptive neuralcontrol method is then putted forward for nonlinear uncertain systems withoutdelay-Generalized Back Propagation (GBP) neural control scheme. This schemeconsidered the nonlinear system to be controlled as a complex nonlinear neuron in theback propagation process. The differential derivative of the output of the controlledsystem with respect to the input can be obtained by sampling the input and output ofthe controlled system periodically, which directly converts the process of neural backpropagation identification process to the control process of the system, and it realizethe unity of recognition and control. The advantage is that the scheme does not need aseparate neural network for model identification of the controlled system or as thesystem observer to predict the system’s conduction; it is a direct adaptive controlscheme. The simulation demonstrates the nonlinear solution control performance ofthis scheme, which has certain reference meanings to the adaptive intelligent controlsystem research of delay free nonlinear uncertain system.4. The feasibility of solving the system identification and control by applying neural network learning techniques is analyzed through several empirical examples,and put forward a method based on Lyapunov stability, which analyzes theidentification and control of nonlinear dynamic system using the comprehensive modelof neural network. In this method, the unknown nonlinear dynamics of the system ismodeled by using neural network modeling technology. The static network structureforms a class of recurrent neural network structure in forms of stable filters combiningdynamic elements, whose structure can approximate a wide class of dynamic systems.A neural network identification and control scheme based on Lyapunov stability isdeveloped, which can guarantee the stability of identification and control schemes evenin the presence of modeling errors. Finally, the proposed scheme is illustrated throughcorresponding simulation examples. Simulation results show that the scheme has goodcontrol performance for a large class of uncertain nonlinear dynamical systems.
Keywords/Search Tags:Nonlinear system, nonlinear control, neural network, neural inverse control, receding horizon optimization technology, adaptive control, BP learning algorithm
PDF Full Text Request
Related items