A Method of Data-driven Parameter Setting Based on Neural Network

A new method of data-driven control parameter setting based on neural network is proposed for the nonlinear controlled objects whose accurate mathematical models are difficult to be established. The design idea is to circumvent the controlled objects and get the controller directly by combining virtual reference and neural network. In addition, the Lyapunov theory is applied to proving that neural network learning rate can guarantee the convergence of tracking error within a certain range. Then the filter of virtual reference feedback tuning (VRFT) algorithm and Taylor expansion are used to further verify the stability of the closed-loop control system. Simulation shows that the method has some advantages of a reduced computational burden, the least amount of calculation, easy parameter adjustment, and strong tracking performance and so on.