Differential evolving to the optimal neural network controller
Primary Area: optimization
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: optimal control, general optimal solution, deep neural networks, Lyapunov dynamics stability, differential evolution algorithm
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Obtaining the general optimal solution for the Optimal Control Problem (OCP) has been a classical but challenging problem for a long time. Existing methods have significant difficulties in addressing such issue and usually OCPs have to be solved numerically with specific boundary conditions given. By leveraging the powerful approximation ability of deep neural networks, a Differential Evolution (DE) algorithm is developed to tackle this challenge, where the deep neural network models are used as the optimal controllers to be optimized. Aiming to the general optimal control problems with free terminal states, which covers a large class of problems, two deep neural networks, namely the Optimal Controller Neural Network (OCNN) and Time Predictor Neural Network (TPNN), are introduced. The dynamic approach is established upon the derived differential evolution equations for the parameters of the neural networks. It is shown that the general optimal solution may be obtained with the proposed method and the resulting neural network controllers perform well as shown by the numerical examples.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1641
Loading