Advances and challenges for scalable cloud-based infrastructure for building data analysis and simulation

Abstract

In the commercial building sector, retro-commissioned and new constructions alike are implementing the use of highly integrated and connected Building Information Systems (BISs) to use fewer resources, improve occupant health and productivity and reduce life-cycle costs. Even if programs such as Leadership in Energy and Environmental Design (LEED) are becoming more and more popular, single factors such as horizontal integration at the portfolio level can drive this change leading to an increase in simulation and forecasting demand. BISs generate large amounts of data from various sources such as control networks or utilities. This data is critical in applications such as continuous commissioning through Automated Fault Detection and Diagnostics (AFDD) or predictive analytics. Moreover, depending on the analysis constraints such as computational maximum runtime or overall cost as well as results' availability and presentation, current techniques can prove to be challenging to use or integrate. This paper aims to introduce to the building performance simulation's space, techniques from other fields such as computer science and data analytics to help improve quality, reproducibility, scalability of workflows used by this industry and research community. Cloud computing and open source technologies described in this paper can help answer many of the aforementioned challenges. When dealing with big datasets, file formats such as Comma Separated Values (CSV) offer little compression resulting in large files creating unnecessary costs and increased query complexity. File formats used by distributed query engines such as Parquet can offer 98% size reduction while being easily queryable using the Structured Query Language (SQL) at the expense of time compared to relational databases which can be subsequently used as a caching mechanism. Containers, often used for cloud-based applications, can also be used for simulation offering scalable, reproducible and hardware agnostic environment to deploy large scale analysis at the expense of a little performance overhead, less than full - fledged Virtual Machines (VMs). Moreover using Infrastructure-as a-Service (IaaS) can allow significant cost reductions by paying only for what is needed in terms of compute and/or memory. For example, running AFDD on 200+ buildings can cost on the order of less than a US dollar per day which is equivalent to a quarter the cost of running a local server. Web technologies such as Javascript-based User Interfaces (UIs) can as well be used to distribute and access simulation results as well as managing the simulation engine altogether remotely. This enables collaboration of international team members as well as deployment of analytics to international or remote clients.