Iterative Feedback Tuning for Regulatory Control Systems Using Identified Sensitivity Functions via Predictive Error Method

The present work proposes the Iterative Feedback Tuning (IFT) for regulatory control systems for single-input, single-output, linear time-invariant (LTI) systems in order to optimize a cost criterion that evaluates the sum of the variance of input and output regulatory control data. The proposed approach estimates the gradient of a cost criterion using identified sensitivity functions at every iteration via the predictive error method (PEM). In the proposed approach, the identified sensitivity function is corrected through the optimization of the sum of squares of one-step predictive error in another regulatory control data with different controller parameters. The correction makes it possible to derive the unbiased estimation of sensitivity function even when the identifiability condition does not hold. Also, the proposed one provides a method for optimization of the step-size of the gradient vector using the identified sensitivity function. A numerical example shows that the proposed one can improve the accuracy of the estimate of sensitivity functions, and achieves faster convergence to the optimal value in the IFT for regulatory control systems.