PI-controller neural tuner appliance to reject disturbances acting on plants of different dynamics

Nowadays, as far as nonlinear industrial plants control is concerned, in most cases the same set of PIcontroller parameters is used both to follow a setpoint schedule and reject disturbances. This set is usually found to solve the first of the mentioned tasks. Not only does that lead to disturbances rejection time increase, but also it increases energy consumption. A neural tuner is developed in this research in order to adjust PI-controller parameters should step-like limited disturbances acting on a plant output emerge. An additional neural network is added to its structure to solve this task. Tuner structure, rule base and algorithm are also improved. The rule base is used to define both moments when to train the neural network and the learning rate values. Experiments are conducted with the help of heating furnace and rolling mill DC drive mathematical models and a real muffle electroheating furnace. Having analyzed obtained results, a conclusion could be made that the tuner usage allows to achieve 20% time decrease to reject disturbances in comparison with a conventional PI-controller. Moreover, energy consumption is decreased by 13.8% for furnace, and amount of rejected products is decreased by 1.5% for the DC drive.