Proper sizing of an HVAC system is essential to guarantee good thermal comfort and achieve high energy efficiency. However, due to the lack of information at the design stage to predict a building’s peak load demand and the lack of operational data to predict the HVAC system operational cost and energy performance, uncertainty exists in the HVAC system sizing process. An incomplete understanding of these uncertainties easily leads to excessive oversizing of HVAC systems, which will inevitably increase the initial, operational, and maintenance costs in the oversized components and causes thermal comfort problems. This thesis provides a systematic study of the uncertainty confronted by the HVAC system design process, particularly in the cooling supply side, and develops a new stochastic sizing strategy for HVAC systems that integrates these uncertainties regarding multiple performance criteria.

In the cooling supply side, the uncertainties in predicting the degradation of chiller capacity are first investigated using the Bayesian Markov Chain Monte Carlo method. Bayesian inference can quantify the uncertainty in aging effects by integrating the prior information and overcoming the problem of data insufficiency. The identified uncertain parameter can describe the aging effects (i.e., the distribution, mean, and ranges) better than the current deterministic evaluation. The calibrated aging parameter has broad applications, such as performance prediction of the HVAC system, decision-making on maintenance schemes, and predicting the remaining service life.

The uncertainties in predicting cooling loss during transmission from the cooling sources (chillers) to the cooling end-users (conditioned zones) are then investigated in the framework of Bayesian inference. The uncertainties in the prediction of the maximum cooling loss are first identified and categorized. The uncertain input parameters are classified into specific and generic types. Sensitivity analysis is conducted to select the important parameters, and the critical parameters are then calibrated using the Bayesian Markov Chain Monte Carlo method. The performance gap between the model prediction and the actual measurement is dramatically reduced by the calibration. The calibration results can inform decision-making on retrofit analysis or used to predict HVAC system performance.

After quantifying the uncertainties in both the load side and cooling supply side, an uncertainty-based HVAC system life-cycle performance evaluation method is developed. The numerical method uses Monte Carlo simulation to propagate the uncertainties from the inputs to the predicted performance indices. Compared with the conventional method of evaluating system performance without considering aging effects or uncertainties, the proposed method provides a longitudinal assessment of the HVAC system performance with more information (i.e., distribution type, mean, and ranges of the predicted performance indicator), which is essential for making rational decisions. A simulation platform is developed based on EnergyPlus for implementation of the proposed life-cycle performance analysis.

Finally, an optimal sizing strategy for HVAC systems is proposed for the purpose of enhancing HVAC system sizing under uncertainties. The optimal sizing strategy defines the alternatives of HVAC system sizes based on the distribution of the predicted peak cooling load, assesses the life-cycle performance of each alternative considering the capacity degradation and cooling loss uncertainty, and selects the optimal design regarding multiple performance indices using the multiple utility theory. The proposed optimal sizing strategy has three advantages over a conventional design. First, it addresses uncertainties directly in the sizing procedure and thus increases the transparency of sizing in relation to risk management. Second, it enables long-term planning and provides more comprehensive information about the system performance for decision-making. Third, it provides a mechanism for integrating the decision-maker’s preferences/requirements on different performance indices into the design, thus facilitating a targeted design that more accurately satisfies the decision-maker’s needs. Case studies are conducted to illustrate the new sizing strategy, and the new sizing method and the old methods are compared.