Introducing a Model-free Adaptive Neural Network Auto-tuned Control Method for Nonlinear SISO Systems

In this study, a novel Adaptive Neural Networks Controller (ANNC) is proposed for controlling single-input single-output nonlinear systems. The proposed ANNC does not rely on an existing model of a system for its weights' training, and does make a full use of the history of the system input and output information for achieving a suitable control effect. The model of the system is used for checking the stability of the system after the calculation of the learning algorithm at each training step, and the controller weights are appropriately tuned to deliver a stable system during the entire training process. Using the accumulated gradient of the system error, the weights' adjustment convergence of the system can be observed and an optimal training number of the system can be selected. The effectiveness of the ANNC in controlling nonlinear industrial plants is demonstrated via simulation. The proposed control scheme provides a building block for the development of comparable schemes useful for more complicated systems involving multiple inputs and outputs.