| Abstract: |
A self-tuning method is an important system design consideration for constructingadaptive controllers of an unknown slowly-varying system. The basic idea in adaptive control isto estimate the uncertain plant parameters and correspondingly adjust control parameters on-line, based on the measured system signals, using the estimated parameters in the control inputcomputation. However, traditional self-tuning adaptive control techniques can only deal withlinear or special nonlinear systems. Typically, these techniques assume that the control model isoperating in a linear region. The parameters of a linearized plant model are estimatedrecursively and used to update the controller. Often, it is not possible to represent, adequately,system characteristics such as nonlinearity, time delay, saturation, time-varying parameters, andoverall complexity. It is important to develop an effective technique in which the structure ofthe nonlinear plant model can be identified by an adaptive process.1.1 Literature ReviewWith emerging development in neural networks, wavelets, andfuzzy logic technologies,adaptive control designs can expand to even greater horizons. Some developments in neuro-controller concept have already proved to be useful for a wide class of practical situations;showing that they can cope with significant unknown nonlinearities [NP90] [LN95]. Theauthors in [KSS91] and [SS91] demonstrated a novel neuro controller using a three-layerdynamical recurrent neural network; however, they assumed that a system state space model ofthe plant was available. The authors in [Che90], [PSY88] and [LS89] employed multilayerfeedforward neural networks in designing an unknown nonlinear self-tuning adaptive controlwith demonstrated potential; however, the backpropagation (BP) training technique IScomputationally complex and usually requires off-line computation to minimize the error.
|
|
|