Neural-net-based model-free self-tuning controller with on-line self-learning ability for industrial furnace

A neural-net-based model-free self-tuning controller for systems with unknown models or some modeling complexity is proposed in this paper. To enhance the on-line self-learning and adaptive abilities, an attenuating excitation signal is introduced to excite all modes of the systems and to produce the error signal needed for self-learning process. To realize the self-organized learning and control, a function evaluating the control effect is introduced to decide whether the on-line operational data can be chosen as the learning samples to train the controller, and how to train. The experiment results for the temperature control problem of some resistance furnaces show the effectiveness of the method