Diseño de un algoritmo automático de sintonización de parámetros de un controlador PID empleando técnicas Bio-inspiradas de aprendizaje de máquinas

Abstract

At present, closed loop control systems mostly used for industrial applications are PIDs, since they have great advantages when compared with other control systems; simplicity of design and low cost of implementation are examples of such advantages. By tuning correctly, the controller manages to increase stability, speed and accuracy in the response. This tuning process presents specific challenges related to the ignorance of several factors (which are raised in section 1) and the few personnel properly trained in the implementation causing a misuse of the technique. This produces a bad tuning of the parameters, carried out through trial-and-error processes or using empirical techniques such as Ziegler Nichols.

Taking into account the previous problems, several companies have focused a large amount of efforts that seek to facilitate the tuning process through algorithms capable of automatically adjusting the parameters of the PID controller. However, the tools developed as a result of such efforts have high associated costs and direct system intervention (undesirable due to downtime in production, decrease in system reliability, etc.).

In this work, we propose the implementation of an intelligent algorithm based on the response of the immune system focused on the principle of clonal selection, whose objective is to find a set of possible solutions that meet design criteria that represent the conditions to which You must adjust the system and the desired result. In the execution of the algorithm the parameters whose results do not satisfy are discarded and those that decrease the error of the figures of merit are replicated. At the end of its execution, the algorithm manages to significantly improve the result of the figures of merit compared with those obtained by the classic tuning technique Ziegler Nichols. The development of the Ziegler Nichols technique is carried out from plant information and shows poor performance in the controller; This process is done manually when performing a simple mathematical analysis. On the other hand, the algorithm performs a heuristic process where various parameters are tested, its response is analyzed and the parameters are refined to obtain a set of solutions that meet the established criteria. This process is performed automatically and may take depending on the processes of information acquisition, presenting a high computational cost with respect to the classical technique chosen, but with a notable difference in the results of the figures of merit.