Predictive control design for large-scale systems

Abstract

The design of a model-based generalised predictive controller (GPC) for large-scale systems is reported. A two-level decentralised Kaiman filter is used to locally estimate the states of each subprocess, and an optimal coordination strategy then improves this filtering solution. A two-level optimisation strategy then decomposes the global GPC problem into manageable subproblems. The GPC solution for each low level subprocess is independently found and sent to update the values of a high-level optimal coordinator. This procedure is repeated until an optimal solution is found. Simulation results for a power generation plant demonstrate the improvement achieved by using the proposed technique.