A data-driven approach for optimizing the reheat process in a variable-air-volume box is presented. Data-mining algorithms derive temporal predictive models from the reheat process data. The bi-objective model formed is solved with a modified particle swarm optimization algorithm. To increase computational efficiency, two levels of non-dominated solutions are introduced while solving the optimization model. A model predictive control strategy is used to generate controls minimizing the reheat output while maintaining the thermal comfort at an acceptable level. (C) 2010 Elsevier Ltd. All rights reserved.