Genetic approach to decentralised PI controller tuning for multivariable processes

A method for the tuning of decentralised PI controllers for multivariable processes, based on genetic algorithms is presented. The power capabilities of genetic algorithms in locating the optimal solutions to a given optimisation problem are exploited by determining the parameters of the PI controllers to meet specified performance objectives. The designer has the freedom to explicitly specify the required performance objectives for a given problem in terms of time-domain bounds on the closed-loop responses. Each set of controller parameters is directly evaluated by simulating the closed-loop system and comparing the resulting responses with the desired using a specially formulated objective function. The proposed method has the flexibility to be applicable to a wide range of multivariable processes. Simulation results illustrate the effectiveness of the proposed method. The choice of genetic algorithms as a suitable optimisation method is supported by comparing them with two conventional optimisation methods