Optimization of an Initial Database using Nelder-Mead Method in designing Database-Driven PID Controller

This paper addresses an optimization issue that the Nelder-Mead (NM) method is implemented to optimize initial database which is used to design Database-Driven (DD) proportional-integral-derivative (PID) controller. As it is well recognized that PID controller, which is considerable applied in industrial processes, is adopted and the parameters tuning are based on DD approach without requirement of model information. The existing DD approach needs to collect a batch of closed-loop operation data, which is used to generate initial database. It is still an open problem that the way of processing data involved in initial database. The proposed scheme considers the NM method, as a free derivative method, to optimize initial database. However, in the evaluation stage, the NM method requires the model information to calculate the output aiming at comparing objective function, therefore, a better vertex can be computed. The fictitious reference iterative tuning (FRIT) is introduced in the proposed scheme to generate fictitious output based on the fictitious reference which is obtained based on input and output data stored in initial database. As a result, the optimized initial database is accomplished, based on which the PID controller then can be designed in Just-in-Time manner. In addition, the efficiency of the proposed scheme is illustrated with a numerical example.