Related Experiment Videos
Trajectory priming with dynamic fuzzy networks in nonlinear optimal control.
Yasar Becerikli1, Yusuf Oysal, Ahmet Ferit Konar
1Department of Computer Engineering, Kocaeli University, lzmit, Turkey. becer@kou.edu.tr
IEEE Transactions on Neural Networks
|September 24, 2004
Summary
Dynamic Fuzzy Networks (DFN) offer a novel approach to optimal control, overcoming limitations of classical fuzzy logic. This self-learning system efficiently generates control trajectories and feedback gains for nonlinear systems.
Area of Science:
- Control Systems Engineering
- Artificial Intelligence
- Computational Intelligence
Background:
- Classical fuzzy logic (FL) faces challenges in dynamic system control, including large rule bases and extended training times.
- Dynamic Fuzzy Networks (DFN) with 'feurons' offer unconstrained connectivity and dynamic processing to address these FL limitations.
- DFNs are explored as a solution for nonlinear optimization problems with dynamic equality constraints in control systems.
Purpose of the Study:
- To introduce and evaluate a Dynamic Fuzzy Network (DFN) as an autotraining system for optimal control trajectory priming.
- To demonstrate DFN's capability in generating optimal feed-forward control trajectories and time-varying feedback gains efficiently.
- To apply the modified-descend-controller (MDC) algorithm and adjoint theory for training DFNs in nonlinear optimal control.
Main Methods:
- A modified-descend-controller (MDC) algorithm, derived from direct-descent-curvature, is employed for nonlinear optimal control computations.
- Adjoint theory is utilized for training the DFN, treating it as a quasilinear dynamic system with reduced computational complexity.
- The Broyden-Fletcher-Goldfarb-Shanno (BFGS) method is used for updating DFN weights and identifying its parameters.
Main Results:
- The DFN-based autotrainer successfully generated optimal control trajectories with fewer iterations compared to traditional methods.
- The algorithm produced robust solutions concerning conjugate points for nonlinear optimal control problems.
- Simulations demonstrated the effective control of a challenging nonlinear second-order system using a three-feuron DFN.
Conclusions:
- Dynamic Fuzzy Networks provide an efficient and effective framework for nonlinear optimal control.
- The proposed autotraining algorithm, incorporating MDC and adjoint theory, significantly accelerates trajectory calculations.
- DFNs offer a promising approach for intelligent control of complex dynamic systems.