| Summary: | 碩士 === 元智大學 === 機械工程研究所 === 82 === When solving the control problem, we often do not know the
exact model of the control plant, or the system may have
uncertainties.The adaptive control theory have been used
extensively, in handling system uncertainties. However, as most
adaptive control theorems are developed based on the as
sumptions of linear model,many existing adaptive control the
orems are not applicable when dealing with systems with large
nonlinearity. Recently, the research on neural network becomes
very popular, and many authors have tried to improve
conventional control algorithms by introducing neural network
in many applications. The objective of this thesis is to use
neural network in the control of input and output data, we can
use neural network to identify the system on line and offline,
and also design appropriate control law based on the learning
result. Opposite to the adaptive control, learning control does
not need the knowledge of system model structure. In addition,
it can be used in many nonlinear applications. In the thesis,
we introduced learning method using back-propagation as well as
associate control design methodology. Examples of several
nonlinear control problems are given with simulated learning
control result. Finally, a single link inverted pendulum was
used to test the learning control methods discussed in the
thesis. The control results demostrate that the neural network
based learning control is an effective method in dealing
nonlinear system with model uncertainties.
|
|---|
