Application-oriented experiment design for industrial model predictive control

摘要: Advanced process control and its prevalent enabling technology, model predictive (MPC), can today be regarded as the industry best practice for optimizing production. The strength of MPC comes from ability to predict impact disturbances counteract their effects with actions, account constraints. These capabilities come use models controlled process. However, relying on a is also weakness MPC.The used by controller needs kept up date changing conditions good performance. In this thesis, problem closed-loop system identification intended in considered.The design experiment influences quality properties estimated model. an application-oriented framework designing used. specifics are discussed. particular, including constraints controllerresults nonlinear law, which complicates design.The time-domain formulated optimal problem, general diffcult solve. Using Markov decision theory, finite state action spaces solved using extension existing linear programming techniques constrained processes. method applies noise disturbance structures but computationally intensive. Two extensions dual implement idea developed. controllers limited output error systems require less computations. Furthermore, since based common technique, they already available implementations. One developed tested extensive experimental validation campaign, first time that propertiesis applied full scale industrial during regular operation plant.Existing procedures most often frequency domain spectrum input variable. Therefore, realization signal right has generated. This not straightforward operating under generating signals, prespecified spectral properties, respect uses ideas stochastic scenario optimization. Convergence desired autocorrelation proved special case merits algorithm illustrated series simulation examples.