Functional enhancements of multivariable predictive controller

The purpose of this Master Thesis concerns further development of ABB´s multivariable prediction controller 3dMPC.The product 3dMPC consists of both on-line and off-line components and controls a multivariable process by a combination of feedback and feed-forward. The Master Thesis involves two assignments: both of them deal with the engineering tool in off-line. One of the assignments was to to make a graphical tool to connect subsystems in the modeling tools in the engineering part of 3dMPC. The model connection algorithm handles cascade connections, i.e. output signals from one model are input signals to another model. State space models for the subsystems are used to create a state space model for the total system. When the total model is constructed, it can be saved to disk as a 3dMPC model file. Assignment number two handles another problem. A controlled multivariable process has normally a number of control signals and some output signals to be controlled, which here are assumed to be pre-defined. In addition to these signals, the 3dMPC has the ability to use other signals to improve the control. There are two types of additional signals: feed-forward signals, which are independent measurable input signals to the process, and measurable output signals from the process, which can be used to improve the estimation of the internal state of the process. In practice is it often hard to know both which measurable disturbances are worth considering, and which additional process variables that could be of any use. To determine the usefulness of these signals, models are identified containing different sets of signals. The scalar norm of the prediction quality of the state space models is calculated and compared. Thus the application assists in the determination of signals and proposes which signals the user should use when building the model to improve the control.Validerat; 20101217 (root