The Research Of Neural Network Identification And Control Of Vibration Systems

This dissertation deals with the demands of intricate vibration control in engineering world, integrates neural network theory and active vibration control, takes the black-box vibration systems as its researching objects, and makes systemic and deep research in intelligent vibration identification and control which bases on artificial neural network technique.As far as the fundament theory and algorithm for system identification and vibration control are concerned, considers the demands of active vibration control, and chooses the BP (Back propagation) algorithm as basic algorithm of vibration system identification and control design. Meanwhile, make deep analysis on mended BP algorithm; a novel dynamic feedforward network model is proposed with local recurrent feedforward features. This model not only breaks free from the inherent limitations in both feedforward and feedback networks, but also gives a full play to non-linear mapping ability of the former and dynamic evolution ability of the latter. T he proposal of dynamic network model and its algorithm make good fundament for neural network identification and control of vibration system.The simulation of vibration system identification and control is accomplished by two MATLAB toolboxes: NNSYSID (Neural Network System I dentification toolbox and Neural Network Control toolbox), mend and renew the network structures and parameters in the toolboxes, a simulink model (or MATLAB M-file) depicts the characters of the objects to be controlled. A MATLAB script file includes the initial variable values and parameters, choose the reference input signals, choose the simple times, simulation iterations, names of model functions, and the parameters design of identification and control systems, etc.Computer simulation results, both MISO and MIMO system identification, or system control, show that methods proposed can gain an effective control over some black-box vibration systems.