Iterative Controller Parameters Tuning Using Gradient Estimate of Variance Evaluation

Data-driven controller tuning approaches directly tune controller parameters using collected data from real systems. In such approaches, the two kind of categories such as non-iterative or iterative approaches have been studied. While the former approaches constitute an alternative cost criterion using collected data, the latter approaches estimate the gradient vector of the original cost criterion. The latter approaches make it possible to achieve the optimal controller gains, but the conventional approaches require two times experiments for the estimation of the cost criterion at given parameters. Especially, the second experiment has to collect the closed-loop response data for a reference signal using the first collected closed-loop response data. For the overcoming the problem, an iterative data-driven controller tuning method for PID controllers was proposed. The proposed approach only requires one time regulatory controlled data for gradient estimation. In the context of the iterative PID gain tuning, the present work extends the approach to the general controller parameter tuning case. While the previous works dealt with the CARIMA (Controlled Auto-regressive Integrated Moving Average) model, and the velocity-type PID controller structure, the present work employs the Box-Jenkins model and a linear parametrized controller structure. Finally, the effectiveness of the proposed approach is shown through a numerical example.