Forecasting the space heating demand using reduced-order modelling approaches for commercial buildings

Energy-efficient retrofits have become crucial in the building sector as approximately 80% of the buildings in developed countries are over 10 years old and consume a major portion of total energy demand. The development and use of thermal models are an integral part of the design process in new and existing buildings due for refurbishment. Building energy performance simulation tools have become quite popular and are now being used to provide estimates of energy consumption at different scales. These tools implement various types of models which differ based on enclosed details. Not all these models are effective in terms of computation and cost. Recently, the total number of developed energy models has grown tremendously, which vary considerably in terms of characteristics and features. Hence, it is crucial to identify the type and characteristics of a model most suited to a certain purpose and situation. Alongside, the sophistication of simulation tools has significantly increased the number of user inputs, thereby, introducing uncertainty in simulation outputs. Grey-box modelling combines the advantages of data-driven and physical modelling approaches. Therefore, these models deliver an appropriate level of accuracy and are also computationally efficient. However, the design approach of grey-box models is often application-specific, for instance, the design approach for grey-box modelling of commercial buildings differs on a case by case basis. Furthermore, the scalability of these models is limited by the network order, which defines the level of complexity incorporated in the model. There is a need for a generalization framework to address the limitations associated with grey-box networks. As different applications require models of varying complexities, it is necessary to identify the model features, such as scalability, flexibility and interoperability, associated with different stakeholders of the building stock. Furthermore, previous uncertainty analysis studies have either failed to segregate the existing uncertainties or considered only one type of uncertainty in the analysis. The thesis introduces a novel generalizable methodology which formulates grey-box networks for different types of commercial buildings. Physical parameters and the nature of the operation of individual buildings constitute the key elements of the developed framework. The framework also relies on the past retrofit history and installed HVAC systems to deduce the order of the grey box network. The methodology further formulates an experimental design technique to associate and assess various model features pertinent to reduced-order grey-box models using pre-defined key performance indicators. Lastly, this research systematically identifies the various sources of uncertainty in grey-box models and develops a framework to include these sources in overall uncertainty quantification. This thesis uses a combination of real-time commercial buildings and reference building archetypes to test and validate the devised techniques. The devised approach reduces the complexities associated with the identification of the network order while maintaining the desired level of accuracy. The approach provides additional insights and information to designers considering novel alternative design approaches, where prior information may not be readily available. The feature assessment frameworks act as a decision support tool in the identification of appropriate model characteristics. The results of this study could support the current need for the assessment of consumption patterns of commercial building stock. The framework could further be implemented to study the post-retrofit heat consumption patterns at the individual building as well as the district scale. A probabilistic framework employing advanced techniques would allow stakeholders to identify influential inputs by considering the factors behind the risks in a given family of distributions.