| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 96 === This thesis presents a Hammerstein-Wiener recurrent neural network with a systematic identification algorithm for identifying unknown dynamic nonlinear systems. The proposed recurrent neural network resembles the conventional Hammerstein-Wiener model that consists of a dynamic linear subsystem embedded between two static nonlinear subsystems. The novelties of our network include: 1) the three subsystems are integrated into a single recurrent neural network whose output is the nonlinear transformation of a linear state-space equation; 2) the well-developed theory of linear systems can be applied directly to linear subsystem of the trained network to analyze its characteristics. To identify a given unknown system efficiently from the input-output measurements, we have derived a systematic identification algorithm that consists of parameter initialization and online stable learning procedures. A frequency domain eigensystem realization algorithm (FDERA) has developed to acquire the system size and to initialize a best-fit state-space representation. To improve the overall identification performance, we first derived an online parameter learning algorithm based on the ordered derivatives. Moreover, to avoid instability of dynamic systems caused by parameter tuning, we have incorporated necessary constraints with the original learning algorithm to form a stable learning algorithm. Finally, computer simulations and comparisons with some existing models have conducted to demonstrate the effectiveness of the proposed network and its identification algorithm. These simulations validate the followings: 1) the proposed network initialization algorithm can provide better initialization than a random initialization approach; 2) with suitable constraints, the proposed stable learning algorithm can ensure the network stability and convergence capability during and after training; 3) the proposed network can closely emulate the behavior of the unknown dynamical system with a satisfactory performance.
|
|---|
