Performance Improvement in Iterative Data-driven PID Gain Tuning Based on Generalized Minimum Variance Regulatory Control

The paper proposes an iterative data-driven control parameter tuning method for Proportional-Integral-Derivative (PID) controllers. The proposed approach uses regulatory control data generated by colored noise, where the reference signal keeps a constant value. In the proposed method, a gradient vector for a cost criterion representing variance of generalized output is estimated using process input and output measurements. The estimated gradient vector is then used for updating PID gains based on a gradient descent approach. The main feature of the proposed method is to estimate the sensitivity function of the closed-loop system from the regulatory control data for estimating a gradient vector of the cost criterion. The approach allows the proposed method to avoid a specific experiment, which is required in the conventional Iterative Feedback Tuning (IFT). Finally, the effectiveness of the proposed method is shown through a numerical simulation.