Abstract
When a feedback control system model is obtained mathematically or physically, a parallel learning system can be composed by copying the model to i (i=1, 2, …, k) systems corresponding to learning times. In the copied systems, the actuating signal of the i- 1-th model is added to the actuating signal of the i-th model. Thus obtained k-th learning model is equivalent to the system which has a filter as series compensator composed by the sum of i (i=0, 1, 2, …, k) multiple of the left side of the characteristic equation. This paper calls the sum "multiple eigenvalue filter", and shows that the multiple eigenvalue filter is applicable in actual plant control systems, to eliminate control variable deviation without losing stability when disturbance is added to the systems. It is demonstrated that control deviation becomes to zero by first learning in the control systems including integral action such as level control. Although controlled systems have an inverse response characteristics which is found in boiler drum level behaviour, the proposed parallel learning method is a powerful way to control the level stably as well as precisely.
