Finite-Time Neural Network Controllers for Nonlinear Dynamical Systems

This paper proposes a simple but novel structure of neural network controllers for nonlinear dynamical systems that achieves finite-time closed-loop control convergence. The proposed structure is motivated by the use of fractional power functions in feedback design, which has proven successful in achieving finite-time convergence. The main contribution of this paper is the use of fractional power functions as activation functions in neural network controllers. Reinforcement learning is used as an example to train the proposed controllers, resulting in optimal finite-time controllers. The effectiveness of the controllers is demonstrated through simulation studies.