Reliably estimating the impact of an active control strategy in a building

Conventional measurement and verification (M&V) methods rely on a pre-/post-intervention analysis that generally take a long time (2 years) to estimate energy savings. Further, results are often affected by changes in building performance that are unrelated to the intervention but that occur over that timeframe. This paper describes a new method which applies to interventions that are easily switchable, such as active control strategies, and which overcomes these limitations. It uses a switchback design that randomly and frequently switches building operations between baseline and intervention strategies at fixed time intervals (e.g., daily). The method uses sequential testing to evaluate pre-defined stopping criteria using measurements obtained by that point in time. Upon satisfying all criteria, the analyst reports savings with an associated uncertainty. The paper demonstrates the method by evaluating a control strategy developed by a software-as-a-service company in a case study building in Chicago, IL, USA. It provides guidance on randomization and blocking, selection of meaningful stopping criteria, implementation of sequential evaluation, and annual weather normalization. The results show that all pre-defined thresholds were satisfied after 45 weeks and the method reliably detected an annual average power savings of 7 kW (±5 kW at 90 % confidence), or 11 % annual energy savings when weather normalized. This result was obtained faster and more robustly than conventional methods, and remained consistent with estimates derived from a further 15 weeks of data. The supplementary material includes example software code to apply this method to other case studies and replicate this analysis. • Addressed drawbacks of conventional measurement and verification methods. • Developed a new method that randomly and frequently switches control strategies. • The method applies to a variety of performance metrics and use cases. • Increases robustness by sequentially testing against pre-defined stopping criteria. • A case study showed improved uncertainty quantification in a reduced timeline.