Robust co-evolutionary design of SISO Smith predictor PID controllers

Genetic algorithms are proposed as a new and novel technique to solve the problem of designing a robust Smith predictor PID controller for a plant with model uncertainties, capable of dealing with time-delays in an exact manner. The evolutionary scheme used, involves generating two separate populations, one representing the controller and the other the plant. The controller population is then co-evolved against a population of plants covering the plant uncertainty search space, such that the controller can control all the plants effectively. A time-domain cost function subjected to a frequency-domain vector margin stability constraint, is then deployed in order to obtain a robust controller design. This evolutionary approach is illustrated by evolving a Smith predictor PID controller for a linear plant which has a set of prescribed model uncertainties, and compares the results with the ones of robust PID control for the same uncertainty operating envelope.