Building energy sector is one of the important sources of energy consumption and especially in the United States, it accounts for approximately 40% of the total energy consumption. Besides energy consumption, it also contributes to CO2 emissions due to the combustion of fossil fuels for building operation. Preventive measures have to be taken in order to limit the greenhouse gas emission and meet the increasing load demand, energy efficiency and savings have been the primary objective globally. Heating, Ventilation, and air-conditioning (HVAC) system is a major source of energy consumption in buildings and is the principal building system of interest. These energy systems comprising of many subsystems with local information and heterogeneous preferences demand the need for coordination in order to perform optimally. The performance required by a typical airside HVAC system involving a large number of zones are multifaceted, involves attainment of various objectives (such as optimal supply air temperature) which requires coordination among zones. The required performance demands the need for accurate models (especially zones), control design at the individual (local-VAV (Variable Air Volume)) subsystems and a supervisory control (AHU (Air Handling Unit) level) to coordinate the individual controllers.

In this thesis, an airside HVAC system is studied and the following considerations are addressed: a) A comparative evaluation among representative methods of different classes of models, such as physics-based (e.g., lumped parameter autoregressive models using simple physical relationships), data-driven (e.g., artificial neural networks, Gaussian processes) and hybrid (e.g., semi-parametric) methods for different physical zone locations; b) A framework for control of building HVAC systems using a methodology based on power shaping paradigm that exploits the passivity property of a system. The system dynamics are expressed in the Brayton-Moser (BM) form which exhibits a gradient structure with the mixed-potential function, which has the units of power. The power shaping technique is used to synthesize the controller by assigning a desired power function to the closed loop dynamics so as to make the equilibrium point asymptotically stable, and c) The BM framework and the passivity tool are further utilized for stability analysis of constrained optimization dynamics using the compositional property of passivity, illustrated with energy management problem in buildings. Also, distributed optimization (such as subgradient) techniques are used to generate the optimal setpoints for the individual local controllers and this framework is realized on a distributed control platform VOLTTRONTM, developed by the Pacific Northwest National Laboratory (PNNL).