Visualization techniques for heterogeneous and multidimensional simulated building performance data sets

The architecture, environment and construction industry is facing, on the one hand, ambitious environmental regulations for low carbon and net zero energy buildings, and on the other hand, the emergence of new techniques such as parametric assessment and cloud computing. As a result, there is a dramatic increase of performance analysis and collected data during the building design phase. However, previous research highlighted major weaknesses of current building performance simulation -BPS- software regarding its ability to represent and explore input and output data, to interact with it, and to extract valuable data patterns and analyses. Therefore, this research aims to identify suitable visualization techniques that might increase the usability and the knowledge extracted from building simulation dataset. To that end, an interdisciplinary approach has been set up. First, a literature review allowed to characterize the specificities of BPS dataset, namely their heterogeneous nature -discrete, ordinal, categorical, and continuous-, their different correlation levels and their medium size. Second, key tasks that should be performed by BPS tools to support the design process are identified: exploration, solutions generation and evaluation. Then, two data visualization techniques that accept the BPS dataset specificities and that enable to perform these key tasks were selected within the information visualization research field: Decision Tree and Parallel Coordinates. Third, these techniques were applied to an extensive BPS dataset, generated from a series of parametric building simulations based on a high-performance building to be, called the smart living building. Finally, a qualitative comparison between the selected visualization techniques was conducted so as to reveal their strengths and weaknesses. This comparison highlights Parallel Coordinates as the most promising approach.