This chapter addresses the problem of controlling a nonlinear process when linear models have been identified at different operating points. It is motivated by the larger problem of transition control, where a controller has to be designed for a plant that operates in multiple regimes and makes transitions between them.
The multiple local models are combined into a single linear parameter varying (LPV) global model. The parameters of the global model are chosen to be model probabilities estimated on-line using a Bayesian approach. A controller is then designed for this LPV structure that is robust in the H-infinity sense.