The Analysis And Simulation Of Model Reference Adaptive Control Systems Based On RBF Neural Network Identifier

Model reference adaptive control(MRAC) is an important adaptive control. It had more mature theories and methods of analysis and synthesis. And in practice, it was becoming used widely, such as autopilot of the aircraft, autopilot system of the ship, servo system of the photoelectric optical tracking telescope, speed-governing system of the controlled silicon, control system of the manipulator and so on.However, the technical means of traditional algorithm for model reference adaptive control is very limited. When it encounter complex nonlinear systems, the design and implementation is very difficult.In order to give full play the superior performance of model reference adaptive control, and increase the robustness, real-time, fault tolerance of control and the adaptive and learning ability of control parameters, the people made model reference adaptive control and neural networks(NNs) up appropriatly to the system of model reference adaptive control based on the neural networks.The thesis analyzed the design of the model reference adaptive control system, and used the radial basis function(RBF) neural network to the nonlinear system identification as an identifier in the control system.Firstly, the thesis introduced the basic theories of the model reference adaptive control system. Then, the thesis introduced three classic design plans of the adaptive control laws and studied the simulation of three classic examples. The thesis analyzed the RBF neural network as an identifier in the control system. Furthermore, the thesis improved the learning algorithm of the RBF neural network and obtained an improved K-means algorithm. Then, a simulation example of system identification had been shown. Finally, the RBF neural network identifier based on improved K-means learning algorithm had been used in the single neuron PID model reference adaptive control system. And a simulation example had been analyzed.The thesis focused on the the simulation and analysis of three kinds of classic design plans of the model reference adaptive control and the summary of advantages and disadvantages. On this basis, the thesis introduced the neural network model reference adaptive control systems and improved the training algorithm of the neural networks identifier. Finally, the thesis gave a simulation example of the neural network model reference adaptive control system.