Parallel Learning in Control Systems : Derivation of Multiple Eigenvalue Filter

Abstract

In this paper, the author proposes a method for improving control performance in feedback control systems by introducing a multiple eigenvalue filter which has been deduced from parallel learning models. First, the control system model is copied to i (i=1, 2, ..., k) systems corresponding to learning times k. The actuating signal of the first model is added to the actuating signal of the second model, and then the actuating signal of the second model is added to the actuating signal of the third model. Likewise, the actuating signal of the k-1-th model is added to the actuating signal of the k-th model. The thus obtained k-th model is equivalent to a system which has a filter as a series compensator that is composed of the sum of i (i=1, 2, ..., k-1) multiples of the left side of the characteristic equation. In this paper, the sum is called "multiple eigenvalue filter" and it is concluded that the filter can eliminate control variable deviation without losing stability when disturbance is imposed.