Self-learning chebyshev fuzzy neural finite time control with application to active power filter

This paper proposes a Self-constructing Chebyshev recursive fuzzy neural network (SCCRFNN) controller using a non-singular terminal sliding-mode controller (NSTSMC) for a class of nonlinear systems. The proposed SCCRFNN structurally combines the Chebyshev neural network (CNN) based on Chebyshev polynomial and the recursive fuzzy neural network (RFNN) to improve the accuracy of a nonlinear approximation, and it also introduces self—constructing algorithm to optimize the structure of neural network. The new proposed SCCRFNN combines the advantages of the two neural networks to achieve a better approximation performance for the nonlinear systems. The main advantages of SCCRFNN are that not only can it handle large-scale problems due to the usage of CNN but also it can adjust the number of hidden layer nodes, which contributes to optimize the overall structure of neural network. And a non-singular terminal sliding-mode controller (NSTSMC) is used with the proposed neural network, which ensure the robustness of controlled system and can converge in a finite time. The proposed SCCRFNN using NSTSMC is utilized to a second-order nonlinear system, that is active power filter (APF) system, to demonstrate the robustness and control performance. A simulation and a hardware experiment are carried out to verify the dynamic property and feasibility of proposed strategy.