A multivariable predictive controller with nonlinear cost minimisation capability has been developed and tested via simulation. The algorithm has been designed for applications where there are more manipulated than controlled variables and each of the manipulated variables has its own associated cost of operation. Basic servo-regulatory behaviour is governed by a feedforward/feedback version of a receding horizon controller. A single user-definable parameter, directly related to the desired settling time of the process, is used to tune the controller. General cost functions are used to synthesize a set of control sequences such that operating costs are minimised without adversely affecting servo-regulatory response. Results are shown for a hypothetical low-order, highly coupled system to illustrate that the algorithm is capable of allocating a proper mix of control action such that responses to setpoint and load changes are swift and smooth and the costs associated with these transient allocations are optimal over the horizon window. Finally, a high-fidelity simulation is used to demonstrate performance in an environmental application where the algorithm is used to control comfort conditions within a building while maximising the efficiency of the space-conditioning plant.

KEYWORDS: calculating, costs, optimisation, optimisers, buildings, predictive controls, testing, algorithms, operations, performance.

Units: SI