An automated approach to developing compact and accurate building models utilising an inverse heat transfer approach

This paper is focused on developing numerical modeling techniques aimed at automating and simplifying the process of generating detailed models for simulating building thermal physics. However, the methodology described can be applied in numerous application areas across the thermals sector. Automated approaches to developing energy models for buildings are of particular importance to alleviate current labor intensive practices of manual generation so that large real-estate portfolios can be analyzed within a reasonable time frame, hence providing a decision making tool that can lead to smarter renovation plans for implementation across the fleet. This process is achieved by developing custom built software code in MathWorks MATLAB® with all aspects discussed herein. The starting point in this analysis is a series of individual CAD drawings for each building in the fleet with spatial coordinates for all lines and nodes loaded into matrices within MATLAB®. A constrained Delaunay triangulation technique is then applied to automatically differentiate building fabric components such as walls, columns, windows, etc., their physical scale, and all interior zones within the building. Multiple floor plans are also automatically linked by layer information, and a series of logical steps are then followed to identify zone-to-zone interactions and exterior building fabrics. A number of generic thermal resistance/capacitance models capable of modeling the anticipated thermal responses of all building elements are defined. These are assigned to the appropriate elements of the building fabric and linked based on the aforementioned zone identification process. The overall result of this process is the automated generation of a thermal energy model for any specific building that is capable of accurately modeling its thermal physics. However, property information specifically relating to on-premise building elements is not automatically discernible from these CAD-based models. In order to address this deficit, typical material properties are assumed for each element of the building fabric in the first instance, and an inverse heat transfer approach is implemented based on a set of limited sensor data for the building in question. This process results in optimizing estimated parameters so that predicted thermal physics match that of measured sensor data. Overall, the paper describes the development of this automated procedure, presents indicative results of its application, and discusses some possible limitations as well as guidelines aimed at alleviating some of these limitations.