下載電子全文宣告This thesis is authorized to indicate in-campus access at 5 years and off-campus not accessible
You can not download at the moment.
Your IP address is 22.214.171.124
The defense date of the thesis is 2012-08-29
The current date is 2019-04-24
This thesis will be accessible at off-campus not accessible
URN etd-0828112-111131 Statistics This thesis had been viewed 845 times. Download 0 times. Author Ming-Hung Chang Author's Email Address No Public. Department Electrical Engineering Year 2011 Semester 2 Degree Ph.D. Type of Document Doctoral Dissertation Language English Page Count 204 Title SELF-CONSTRUCTING FUZZY NEURAL NETWORK AND DIFFERENTIAL EVOLUTION ALGORITHM FOR IMPLEMENTATION OF PARAMETER ESTIMATION Keyword Self-constructing fuzzy neural network modified differential evolution algorithm. rule elimination rule generation adaptive law robust controller Mahalanobis distance PID controller PID controller Mahalanobis distance robust controller adaptive law rule generation rule elimination modified differential evolution algorithm. Self-constructing fuzzy neural network Abstract The fuzzy neural network (FNN) possesses the advantages of both fuzzy logic process and neural network; it combines the capability of fuzzy reasoning in handling uncertain information and the capability of neural networks in learning from process. For a tracking control, unknown parameter and uncertain external perturbation causes the nonlinear system to be unstable. In this dissertation, we apply the adaptive self-constructing fuzzy neural network (ASCFNN) to real-time estimate the parameters of the real nonlinear system. The initial structure of ASCFNN has input layer and output layer. The hidden layers are generated automatically and dynamically in the learning process according to the online incoming data by performing the structure learning process. The Mahalanobis distance (M-distance) method in the structure learning is also employed to determine if the hidden layers are generated/ eliminated. Concurrently, the adaptive laws are derived based on the sense of Lyapunov so that the stability of the system can be guaranteed. In order to compensate the uncertainties of the system parameters and achieve robust stability of the considered system, the robust controller is adopted. Due to the fixed parameters, the traditional Proportional-Integral-Derivative (PID) doesn’t have the capability of learning for the change of external disturbance of the nonlinear system. In this dissertation, the ASCFNN is employed to estimate online and update the parameter of the traditional PID controller to achieve better tracking performance.
Differential evolution (DE) algorithm is an efficient and powerful population-based stochastic search technique for solving optimization problems over continuous space, which has been widely applied in many scientific and engineering fields. DE algorithm has been employed as a robust optimization algorithm and successfully applied to solve various difficult optimization problems. For parameter learning of FNN, employing a trial-and-error scheme to search for learning rates of the Back-Propagation (BP) algorithm and its associated parameter settings require high computational time. In this dissertation, the system dynamic model is substituted into DE estimator to estimate offline the preset parameters of the FNN identifier. After the best preset parameters are obtained, the time varying system is online controlled by using FNN identifier.
Finally, the simulation results and actual experiment are implemented to demonstrate robustness, effectiveness and accurate tracking performance of the proposed controller under the conditions of external disturbance.
Advisor Committee Hung-Ching Lu - advisor
Chih-Min Lin - co-chair
Teng Pin Lin - co-chair
Tzuu-Hseng S. Li - co-chair
Ya-Fu Peng - co-chair
Files Date of Defense 2012-07-04 Date of Submission 2012-08-29