期刊
IEEE CONTROL SYSTEMS LETTERS
卷 5, 期 2, 页码 587-592出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/LCSYS.2020.3004747
关键词
Derivative-free optimization; exploding and vanishing gradient; data-driven systems; linear quadratic regulator
资金
- Office of Naval Research [N00014-19-1-2556, N00014-19-1-2266]
- Air Force Office of Scientific Research [FA9550-191-0283]
- NSF [DMS-1847802, ECCS-1829821, CNS-1828678, SAS-1849228, GCR-1934836]
- NRL [N00173-17-1G001, N00173-19-P-1412]
- NOAA [NA16NOS0120028]
In this letter, a bio-inspired derivative-free optimization algorithm is proposed for minimizing objective functions with vanishing or exploding gradients. The algorithm shows fast convergence and robustness to noisy function evaluations, and has applications in finding LQR gains for unstable systems.
In this letter, we propose a bio-inspired derivative-free optimization algorithm capable of minimizing objective functions with vanishing or exploding gradients. The proposed method searches for improvements by leveraging a PCA-based strategy similar to fish foraging. The strategy does not require explicit gradient computation or estimation and is shown in simulation to require few function evaluations. Additionally, our analysis proves that the proposed algorithm's search direction converges to the gradient direction everywhere outside of small neighborhoods around local minima. Applications to a data-driven LQR problem and noisy Rosenbrock optimization problem are demonstrated. Empirical results show the proposed method exhibits fast convergence and is able to find the LQR gains for any controllable system, including unstable systems, and is robust to noisy function evaluations.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据