The objective of the research is development of practical, multivariate model predictive control (MPC) for fossil power plants. The approach used here involves the development of a generic, reduced-order model (ROM) which captures the dominate static and dynamic characteristics of power plants. This model is necessary in order to predict the plant behavior. Since the model will not exactly match the true plant structure, the parameters of the model must be estimated using prediction error methods or nonlinear least squares. This model will then be used in a Kalman filter to estimate process states in real-time. These estimated states used for prediction, enabling the computation of the optimal control sequence. The research builds on the results of a preliminary study which attempted to demonstrate feasibility of the approach. The results were encouraging but much more work is needed to make the approach practical. In particular, the current structure of the ROM does not adequately model the behavior of a plant and thus more modeling work is required. Another problem is the computational burden, which currently precludes real-time operation on a microcomputer. Several approximation techniques can potentially reduce the burden by a factor of ten. Finally, parameters of the MPC algorithm (weighting and control step size) must be evaluated to achieve the best performance. The successful application of the method will enable increased efficiency and availability, and reduced maintenance and pollution.
