Augmenting building performance predictions during design using generative adversarial networks and immersive virtual environments

Highlights

• A framework combining context-aware design-specific data with an existing building performance model (existing BPM) guided by a performance target was proposed.

• An immersive virtual environment (IVE) simulating a building under design was used to acquire context-aware design-specific data.

• A generative adversarial network was used to combine an existing BPM with context-aware design-specific data using a performance target as guided.

• A case study demonstrated efficacy and reliability of the framework.

Abstract

Existing building performance models (existing BPMs) often lack the capability for addressing human-building interactions in future buildings or buildings under design because they are mainly derived using data in existing buildings. The limitation may contribute to discrepancies between simulated and actual building performance. In a previous study, the authors discussed a framework using an artificial neural network (ANN)-based greedy algorithm which combines context-aware design-specific data obtained from immersive virtual environments (IVEs) with an existing BPM to enhance the simulations of human-building interactions in new designs. Although the framework has revealed the potential to improve simulations, it cannot determine the appropriate combination between context-aware design-specific data and the existing BPM.

In this paper, the authors present a new computational framework (the GAN-based framework) to determine an appropriate combination based on a given performance target to achieve. Generative adversarial networks (GANs) are used to combine data of an existing BPM and context-aware design-specific data using a performance target as a guide to produce an augmented BPM. The effectiveness and the reliability of the GAN-based framework were validated using an IVE of a single occupancy office. Thirty people participated in an experiment on the simulation of artificial lighting switch uses using the IVE. Their light switch uses data under different work area illuminance were collected and analyzed. The building performance models (BPMs) proposed by Hunt and Da Silva were selected as the existing BPM and the performance target respectively. The data of each participant was used to generate an augmented BPM using the GAN-based framework and an updated BPM using the previous framework (i.e., ANN-based greedy algorithm framework). The thirty pairs of the augmented and updated BPMs were compared. Specifically, the errors measured between the updated BPMs and the performance target (E₁) and the errors measured between the augmented BPMs and the performance target (E₂) were analyzed using t-tests (α = 0.05). In 22 out of 30 cases, the performance of the augmented BPMs was significantly better than the updated BPMs, and in four cases, the performance of the two was similar. Only in four other cases, the performance of the updated BPMs was better. The results confirmed the efficacy of the framework. However, future research is needed to study the performance target and uncertainties associated with IVE experiments to better understand and control the reliability of the framework.

Introduction

The design stage of a building project is a critical step to make decisions and establish directions for engineering building components, affecting the characteristics, functions, and performance of a building. To optimally translate design goals and objectives into the performance of a building, designers and engineers usually apply building performance models (BPMs) during the design stage such as simulations of building energy consumptions and human-building interactions to understand, investigate, and predict building performance, as well as support decision-making. Nevertheless, the application of BPMs cannot eliminate the significant performance discrepancies between the simulated and the actual performances that have been widely reported [[1], [2], [3]]. For example, studies have reported as much as 150% of differences between predicted and the actual performance of a building [4].

Many factors influence the simulations of building performance, especially human-building interactions such as occupant responses to building contexts and occupant habitual behaviors [5]. Human-building interactions are highly context-depended and sensitive to several contexts [6,7] in which contexts are described by situational factors that are not directly included in a model or simulation [8]. These situational factors are often assumed to remain constant across different applications of the model or simulation. For instance, “context” may be physical or natural factors (e.g., building characteristics, building surrounding and environmental factors, and climate conditions), and socio-technical factors (e.g., participant's cultural background, racial/ethnicity, and tasks to perform), which may not be included as variables in a BPM. However, such factors can have an impact on analysis using the BPM during the design of a specific space, whose situational factors may be different from what the BPM has assumed and cannot be treated as constant across different applications. In such cases, these situational factors in relation to any BPM need to be identified, analyzed, and integrated in building performance analyses. Often, BPMs are developed using data obtained from existing buildings, where the contexts of which differs from the contexts of a building under design. Applying such BPMs to understand, investigate, and predict human-building interactions in a building under design may contribute to the discrepancy between predicted and actual performance. Therefore, being able to address human-building interactions responding to specific contexts in new designs (e.g., the context embodied) can potentially enhance the accuracy of BPMs leading to reductions of the discrepancy between predicted and actual performance of a building.

Immersive virtual environments (IVEs) have demonstrated their potential in simulations and data collections in many disciplinary areas, especially engineering fields such as emergency evacuations [9,10], building designs [11], and human-building interactions [[12], [13], [14]]. IVEs provide several advantages over other data collection methods such as sensing, field studies, and surveys. For instance, IVEs can replicate certain context for buildings under design, especially when the contexts cannot be possibly, cost-effectively, or safely replicated in reality. Additionally, IVEs allow users to fully handle experimental conditions, and customize experimental models as desired. Human-building interactions in buildings under design may not be directly observed and analyzed. As a result, the application of IVEs can be an alternative for generating and examining the context-aware design-specific data of a new design. Following Sowa's definition of context [8], “context-aware” refers to the capability of a method, simulation, or model to address the impact of identified contextual factors in analysis. Therefore, by using the method, simulation or model, users are able to consider human-building interactions responding to contexts of a specific design. For example, in the application discussed in the paper, the context-aware design specific data of the proposed computation framework included contextual factors such as types of office task and locations of light switch.

To improve the accuracy of existing building performance models (existing BPMs), the authors have offered a framework for customizing existing BPMs to address contextual factors of a building under design. The framework using an artificial neural network (ANN)-based greedy algorithm has been developed to combine an existing BPM with context-aware design-specific data obtained from IVE experiments [15]. The framework has shown the potential to enhance the prediction accuracy of an existing BPM. However, its major limitation is that it lacks the capability to determine the appropriate combination of an existing BPM and context-aware design-specific data in a principled way rather than through trial and error, that can entail excess resource and time consumption. Hence, the principal goal of this study is to improve the capability of the framework to be able to determine the appropriate combination without trial and error. The new computational framework applies generative adversarial networks (GANs) to combine an existing BPM with context-aware design-specific data obtained from IVE experiments, and uses a performance target as a guide during computation to determine the appropriate mix without trial and error. The GAN-based framework produces an augmented building performance model (an augmented BPM) representing the appropriate combination that satisfies the performance target.

In the following, the authors first discuss comparison of the GAN-based framework and the ANN-based greedy algorithm framework, and then provide the research objective followed by an expression of the GAN-based framework and the explanation of applying the framework on a single-occupancy office to validate the framework. The design and administration of the IVE experiment are explained in detail. Finally, results, discussions, and limitations of the study, as well as conclusions and directions of future work are provided.