Genetic PI Controller Tuning to Emulate a Pole Assignment Design

Abstract

Tuning of proportional-integral (PI) and proportional-integral-derivative (PID) controllers continues to be a current topic, as the control needs in industry are broad and diverse. Although PID controllers have been the predominant controller type for several decades, there are still opportunities to improve the performance of both the installed base and newly deployed controllers. One of the main challenges is a reliable and easy tuning that can be performed by an operator on site. From a computer science point of view, the problem of tuning PID controllers qualifies as a nondeterministic polynomial-time hard problem (NP-hard). Genetic algorithms are a heuristic approach to approximate this type of problem, and the continued growth of computational capacity makes them increasingly more viable. In this work, we present a PID tuning architecture using a genetic algorithm, which incorporates in its fitness function an emulation of pole assignment and implicit cancellation of the additive dynamics of zeros in a closed loop, in addition to reducing discretization losses. The tuning is performed with a paradigm different from the one usually applied in the literature. Rather than simply minimizing the error between the process variable and the setpoint, the error between the process variable and an ideal response curve associated with a desired pole assignment is minimized. This approach provides better control over closed-loop performance.