Probabilistic Methods and Big-Data for Risk-Robust Building Systems

This dissertation explores the problem of making building design decisions when facing complex systems with interactive effects and unreliable or inadequate information. I approach the understanding of decision making from a probabilistic point of view, with an emphasis on utilizing increasingly-available measured data. In particular, I encourage the characterization of uncertainty in predictions of building behavior and the use of the estimates to understand and weigh risks.Intelligent design and operation of building systems can significantly reduce operating costs, mitigate environmental impacts, and minimize occupant health effects. To do so, building systems must be understood from a probabilistic point of view, i.e., the relationship between system design and the likelihood of improving building performance must be characterized. The availability of measured data on building systems and performance has grown in recent years, and is likely to continue growing. These data provide an opportunity to understand design trade-offs, but data are often noisy and incomplete. Realizing their full utility requires statistical models that can quantify uncertainty. Probabilistic methods are under-utilized in the field of building systems; analytical and theoretical models are often used instead. Thus, this dissertation focuses on understanding building systems by utilizing probabilistic techniques informed by measured data.In Chapter 2, I present a probabilistic approach to designing an indoor sampler network for detecting an accidental or intentional chemical or biological release, and demonstrate it for a real building. I develop an algorithm to design sampling architectures which maximize the probability of detecting a release, and which minimize the time to detection. Using a model of a real, large, commercial building, I demonstrate the approach by optimizing networks against uncertain release and sampler characteristics. Finally, I speculate on rules of thumb for general sampler placement.In Chapter 3, I present methods for quantifying uncertainty in predictions of baseline building energy use. I show that uncertainty estimation improves measurement and verification (M\&V) information and overcomes some of the difficulties with deciding how much data is needed to confirm energy savings. I show that cross-validation is an effective method for computing uncertainty, and extend a regression-based method of predicting energy use using short-interval meter data. I demonstrate the methods by predicting energy use in 17 real commercial buildings. I discuss the benefits of uncertainty estimates which can provide actionable decision making information for investing in energy conservation measures.In Chapter 4, I demonstrate an approach to estimating energy savings due to implementing building equipment retrofits. I show that building data and statistical algorithms can provide savings estimates when detailed energy audits or simulations are not cost- or time-feasible. I develop a multivariate linear regression model to quantify the contribution of building characteristics and systems to energy use, and use it to infer the expected savings when modifying particular equipment. I apply the model to a large collection of building data. I discuss the ways understanding the risk associated with retrofit investments can inform decision making.The scientific contribution of this dissertation is a new probabilistic approach to designing and operating building systems. I provided a clearer understanding of the risks associated with their design and operation. I present methods for utilizing noisy and incomplete data to design systems that are robust with respect to the uncertain conditions in which they must operate