Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Intelligent optimal control with dynamic neural networks.

Yaşar Becerikli1, Ahmet Ferit Konar, Tariq Samad

  • 1Department of Computer Engineering, Kocaeli University, Izmit, Turkey. becer@kou.edu.tr

Neural Networks : the Official Journal of the International Neural Network Society
|March 12, 2003
PubMed
Summary

Dynamic neural networks (DNNs) enable efficient intelligent optimal control by overcoming limitations of traditional architectures. This self-learning system generates optimal control trajectories and feedback gains faster, paving the way for real-time applications.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Automatic identification of landmarks for standard slice positioning in brain MRI.

Journal of magnetic resonance imaging : JMRI·2011
Same author

Nonlinear predictive control of a drying process using genetic algorithms.

ISA transactions·2006
Same author

Intelligent adaptive nonlinear flight control for a high performance aircraft with neural networks.

ISA transactions·2006
Same author

Direct descent curvature optimal control with a modification algorithm and its application to nonlinear process.

ISA transactions·2006
Same author

Trajectory priming with dynamic fuzzy networks in nonlinear optimal control.

IEEE transactions on neural networks·2004

Area of Science:

  • Control Engineering
  • Artificial Intelligence
  • Computational Neuroscience

Background:

  • Traditional neural network architectures face limitations in dynamic system control, including large network sizes and long training times.
  • Dynamic Neural Networks (DNNs) offer a potential solution to these challenges by incorporating dynamic properties into network design.

Purpose of the Study:

  • To investigate the use of DNNs for intelligent optimal control problems, specifically framing it as a nonlinear optimization with dynamic equality constraints.
  • To develop a self-learning algorithm that utilizes DNNs as a control trajectory priming system.

Main Methods:

  • An auto-training algorithm for DNNs was developed, functioning as a self-learning structure.
  • Optimal control computations were performed using a modified direct-descent-curvature algorithm (modified-descend-controller-MDC).

Related Experiment Videos

  • Adjoint theory was employed for DNN training, treating it as a quasi-linear dynamic system, with weight updates based on the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method.

    • Main Results:

    • The DNN-based approach successfully generated optimal feed-forward control trajectories and time-varying optimal feedback gains.
    • The algorithm significantly reduced the number of iterations required for computation.
    • The modified-descend-controller-MDC algorithm demonstrated numerical robustness concerning conjugate points.
    • Simulations were conducted on a challenging nonlinear second-order system using a three-neuron DNN.

    Conclusions:

    • Dynamic Neural Networks provide an effective method for intelligent optimal control, overcoming the limitations of static architectures.
    • The developed self-learning algorithm accelerates trajectory calculations, enabling real-time intelligent optimal control with virtual global feedback.
    • The approach demonstrates the potential for encapsulating and generalizing optimal control trajectories within DNNs.