Coordinated-distributed optimal control of large-scale linear dynamic systems

Abstract

Since the late 1970s, the design of chemical processes has evolved towards highly integrated operations that can increase plant efficiency. Although plant-wide integration offers substantial opportunities to improve the performance of an entire plant, it often results in large-scale systems that are challenging to control. Distributed control has attracted increasing attention in recent years and is seen as a promising new strategy for control of large-scale systems. This thesis focuses on a class of distributed control schemes, which is referred to as 'coordinated-distributed' control. The coordinated-distributed control schemes proposed in this work use a coordinator that exchanges information with local controllers. This exchange of information allows the coordinator to iteratively adjust (coordinate) the optimal control problem for each of the local subsystems to drive their solutions toward the plant-wide optimal performance operations. In this work, coordinated-distributed control schemes are formulated for dynamic linear systems that can be locally controlled by linear quadratic controllers or model predictive controllers. Two distinct methods are used for coordination of the local controllers: the prediction-driven coordination method and the price-driven coordination method. A common characteristic in both coordination methods is the computation of a price vector. The price vector is updated iteratively by the coordinator to achieve the desired plant-wide optimal performance. The coordinated-distributed control schemes proposed in this work assume the same execution rate for the local controllers. An extension for coordination of local controllers that are executed at different rates is also presented. Different strategies for dual-rate coordination are discussed and the one that can provide the most performance improvement is analyzed in detail. The coordinated-distributed control strategies proposed in this thesis have significant potential for improving the performance of current industrial process control systems. The proposed coordinated-distributed control schemes do not require a radical new configuration of the decentralized controllers. They can be constructed with minor modifications to the existing decentralized control systems.