طراحي كنترل كننده شبكه موجك فازي تطبيقي به روش غير مستقيم براي دسته اي از سيستم هاي غير خطي

چكيده انگليسي :

123Indirect Adaptive Fuzzy Wavelet Network Controller Design for a Class of Nonlinear Systems Maryam Shahriari kahkeshi m shahriyarikahkeshi@ec ut ac ir 2010 10 19 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisors Farid Sheikholeslam sheikh@cc iut ac irMaryam Zekri mzekri@cc iut ac irAbstract Performance of many physical systems is inherently nonlinear and must be described by nonlinearmathematical models But some of these systems have unknown structure and it is not possible to provideaccurate mathematical model for them Thus conventional control methods cannot be used to control thesesystems Therefore intelligent computational techniques such as fuzzy logic neural networks geneticalgorithms and so on are recently applied to solve control problems of dynamic systems with unknownstructure or systems with uncertainty in the structure and parameters On the other hand in recent years based on combination of intelligent computation and wavelettheory new methods such as wavelet neural networks WNNs and fuzzy wavelet networks FWNs hasbeen proposed Because these networks combining neural network learning ability and properties ofwavelet function not just preserve the multi resolution analysis of wavelet but also have the advantagessuch as simple structure high approximation accuracy and a good generalization capability for nonlinearsystems In this thesis an indirect adaptive fuzzy controller for a class of nonlinear systems based on fuzzywavelet networks is presented Proposed controller uses two fuzzy wavelet neural networks to approximateunknown dynamic of system in off line operations At first based on training data and by constructingwavelet lattice candidate wavelets are selected Then by using OLS algorithm effective wavelets areselected among candidate wavelets to construct sub wavelet neural networks sub WNNs Each fuzzy rulein fuzzy wavelet network structure correspond to one sub wavelet neural network and one adaptationparameter Each sub wavelet neural network is consist of wavelets with specified dilation parameter Thus with prescribing the number of fuzzy rules and sub wavelet neural networks structure of each fuzzywavelet network to approximate system dynamic is determined In the next step extended Kalman filter EKF algorithm and recursive least square estimator RLSE are used for tuning network parameters sothat fuzzy wavelet network achieve an approximation with high accuracy and fast convergence Therefore at the end of off line step the structures of each fuzzy wavelet network to approximate system dynamic isobtained In the next step these structures are used to approximate unknown system functions and adaptivelaws for tuning adaptive parameter in each fuzzy rule are designed based on Lyapounov approach so thatthe closed loop system be stable and tracking error converges to zero In order to illustrate the efficiencyand ability of proposed control algorithm simulation results for nonlinear system servomechanism andinverted pendulum are presented Key wordsFuzzy wavelet networks adaptive fuzzy control nonlinear control wavelet neural networks

شهرياري كاهكشي، مريم

طراحي كنترل كننده شبكه موجك فازي تطبيقي به روش غير مستقيم براي دسته اي از سيستم هاي غير خطي

شبكه هاي عصبي موجك