Investigation of Energy Modelling Methods of Multiple Fidelities: A Case Study

Abstract

Building energy modelling has become an integral part of building design due to energy consumption concerns in sustainable buildings. As such, energy modelling methods have evolved to the point of including higher-order physics, complex interconnected components and subsystems. Despite advances in computer capacity, the cost of generating and running complex energy simulations makes it impractical to rely exclusively on such higher fidelity energy modelling for exploring a large set of design alternatives. This challenge of exploring a large set of alternatives efficiently might be overcome by using surrogate models to generalize across the large design space from an evaluation of a sparse subset of design alternatives by higher fidelity energy modelling or by using a set of multi-fidelity models in combination to efficiently evaluate the design space. Given there exists a variety of building energy modelling methods for energy estimation, multi-fidelity modelling could be a promising approach for broad exploration of design spaces to identify sustainable building designs. Hence, this study investigates energy estimates from three energy modelling methods (modified bin, degree day, EnergyPlus) over a range of design variables and climatic regions. The goal is to better understand how their outputs compare to each other and whether they might be suitable for a multi-fidelity modelling approach. The results show that modified bin and degree day methods yield energy use estimates of similar magnitude to each other but are typically higher than results from EnergyPlus. The differences in the results were traced, as expected, to the heating and cooling end-uses, and specifically to the heat gain and heat loss through opaque (i.e., walls, floors, roofs) and window surfaces. The observed trends show the potential for these methods to be used for multi-fidelity modelling, thereby allowing building designers to broadly consider and compare more design alternatives earlier in the design process.

Multidisciplinary concurrent optimization framework for multi-phase building design process

“…In similar situations, engineering design fields utilize metamodels (simple approximations of higher fidelity models) to explore large sets of design alternatives at a lower computational cost (Alexandrov et al, 2001;Simpson et al, 2001;Fernández-Godino et al, 2016). In the AEC field, there has been a recent surge in metamodels for structural analysis (Brown and Mueller, 2016;Chhabra and Warn, 2018), energy (Tresidder et al, 2012;Muthumanickam et al, 2018;and daylighting (Wortmann et al, 2015;Ayoub, 2020;Muthumanickam et al, 2022a). Given such developments, it might be possible to strategically use metamodels of increasing fidelity across multiple phases of building design as a means to identify the optimal solutions at a lesser computational cost.…”

“…Given such developments, it might be possible to strategically use metamodels of increasing fidelity across multiple phases of building design as a means to identify the optimal solutions at a lesser computational cost. To that end, this paper illustrates strategies to organize energy and daylighting metamodels of varying fidelities (proposed in Muthumanickam et al, 2018Muthumanickam et al, , 2022a across multiple phases of design optimization of an office building. Specifically, a modified bin method and a degree day method (uses lower-order physics equations for heat balance calculation than sophisticated tools like EnergyPlus™) and EnergyPlus™ were used for energy estimation in the example.…”

“…Specifically, a modified bin method and a degree day method (uses lower-order physics equations for heat balance calculation than sophisticated tools like EnergyPlus™) and EnergyPlus™ were used for energy estimation in the example. A detailed overview of the differentiating features of the modified bin method, the degree day method, and EnergyPlus™ that makes them models of varying fidelities is covered in Muthumanickam et al (2018). Similarly, an artificial neural network (ANN)-based metamodel trained using input-output dataset from higher fidelity tool -RADIANCE™, Diva-for-Rhino™, and native RADIANCE™were used for spatial daylight autonomy (sDA) estimation, in the example.…”

“…The modified bin, degree day methods, and the ANN-based daylighting metamodels cited in the section "Multi-phase concurrent coupled MDO framework" showcase promising potential to help evaluate large sets of building designs for energy and daylighting at a lower computational expense. While the working mechanisms of these multi-fidelity metamodels are extensively covered in Muthumanickam et al (2018) (metamodels for energy) and Muthumanickam et al (2022a) (metamodels for daylighting), this paper focuses more on optimal organization and usage of these metamodels across multiple phases of building design to ensure that optimal solutions are preserved at a relatively lower computational expense and time than classical simulations. It should be noted that by using a variety of such metamodels to evaluate large catalogs of generated design options, more amount of design data is generated.…”