ABSTRACT
Model Predictive Control (MPC) algorithms make use of predictions of the system behaviour. This chapter shows how to compute those predictions for various types of linear models. For linear models, the predictions of future outputs are affine in the current state and the future control increments. The Controlled auto-regressive integrated moving average model is first replaced by its incremental form, as this allows offset free prediction in the steady state and also one can ignore the unknown term. Finite impulse response (FIR) models are the most common models utilised in commerical MPC packages. The algebra is far easier than with state-space models and transfer function models because there is no auto-regressive part in the model; predictions depend solely on the input information. When using independent model for prediction, it is equivalent to the use of an FIR model with no truncation errors; this is because the prediction is based solely on input information.
