| Summary: | 碩士 === 國立臺灣大學 === 化學工程研究所 === 81 === The ignorance of the process's state information and the
unmodeled disturbance are the main effects which rise the
perbutation of an observer system. In this thesis, we perform
a it initial state predictor which constructed by a neural
networks and it can predict the current state by using the
current and priori information of the measurable input/output
data. Moreover, a new observer design algorithm based on the
Lyapunov linearization method is presented. Combing the
initial state predictor to the observer system, all the state
variables and unmodeled disturbance are estimated. With the
stability analysis, a constrain of choosing observer pole is
declared and guarantees the uniformly asympotical stablility of
the observer. Simulation results show that the proposed method
is less restrictive than other method and able to obtain
resonable performance for state estimation and unmodeled
distburance.
|
|---|
