Supervisory Control of a Commercial Heating, Ventilating, and Air Conditioning System

Abstract

The MPC supervisory controller generated 'optimal' setpoints for lower-level single-input, single-output controllers to maximally utilize the control system framework currently used in the industry. Two different supervisory controller configurations were developed and compared with the existing normal mode of control. Results showed that superior performance could be obtained using the model predictive supervisory controller but this required non-trivial knowledge for proper identification and tuning adjustments. The framework of MPC was shown to be well suited for supervisory control of HVAC processes due to the inherent characteristics of the algorithm. The reformulation of MPC to use the compact, state-space models with optimal state estimates generated by the subspace identification algorithm offers promise for improved measurement and disturbance characterization and future automation of the identification process.