運用データに基づくオブザーバ構造型制御器設計における計算複雑さ削減

— データのダウンサンプリングによる計算複雑さ削減と推定性能のトレードオフ例 —

抄録

The effectiveness of observer-based controllers using Luenberger observers is well recognized. As the estimated states can be used for possible fault detection, health monitoring, etc., the estimation of plant states is useful. Therefore, a conversion method of transforming a priori designed controllers to observer-based controllers has been proposed. This conversion method is attractive, but it is applicable only for the case that LTI plant systems are controlled by LTI controllers. To enhance the applicability, observer-structured controllers with a structural similarity to the Luenberger observer-based controllers have been proposed, and the conversion method to observer-structured controllers using state-transformation matrices has also been proposed. This method uses mathematical models instead of operational data, thus, observer-structured controllers inevitably suffer from plant state estimation degradation due to unforeseen uncertainties and disturbances in real environments. On this issue, under the condition that a few sets of the plant and controller states in real operating environments are available, a method for obtaining state-transformation matrices using the obtained operational data has been proposed with an illustrative example. However, there is a concern that the numerical complexity in designing state-transformation matrices increases as the amount of the data grows. To solve this concern, this paper proposes a remedy of numerical complexity reduction in obtaining state-transformation matrices using operational data with downsampling approach. The usefulness and the validity of our method are confirmed through a numerical example.