Practical challenges towards data-driven applications in buildings: lessons-learned from two real-life case studies

Data-driven applications in buildings using AI and machine learning have generated a lot of interest, but scaling these applications is challenging due to the uniqueness of each building, including different systems and sensors, lack of standardized meta-information, loss of data, and errors in labels and metric units. Based on two real-life case study, six major challenges for real building implementation of data-driven services using predictive control as an example are outlined, detailed examples for these respective challenges are provided and seven take-home messages for the reader with regards to application and development of data-driven services are proposed: i) The application of data-driven services is not trivial as each building is different, ii) Malfunctioning devices, stable communication and limitations of communication protocols are a major technical barrier for data-driven applications, iii) Know what you measure!, iv) Data-quality is of outmost importance, v) Integration can be complex, vi) Think scalability and vii) Identify user-friendly solutions that are easy to use and understand.