Multivariable controller tuning by genetic algorithms

Multivariable control design methods usually involve quite complex procedures and algorithms. The approaches which try to avoid this complexity can be found between the so called tuning methods. The simplification can be explained through the following ideas: the chosen structure is usually PI-controller, the complete model information is not an imperative, tuning procedure is comparable with the methods for univariable systems, this. Means for the systems with single input and single output. The potential difficulty which can in such situations be expected is that the resulting closed-loop structure is far from an optimal. That means that retuning of the parameters only (additional design freedom) can result in much better solution. In the paper the procedure is suggested for optimisation approach where genetic algorithms are combined with some other design approaches. Specified criterion is rather general and can be regarded as a part of design procedure but it can also be used as selfstanding design approach. The efficacy is presented through results obtained for the model of the real pilot plant of three coupled tanks.