Automatic detection of control loops in existing buildings

Abstract

Improving control loops in buildings is cost-intensive due to the need of expert time. With well structured data, the development of data-driven algorithms is promising to minimize the required time. But data stream names across buildings do not follow a uniform naming schema.We developed a method to identify temperature control loops (sensors and set points) based on data stream names. The methodology starts with narrowing down to three potential sensors per set point using string distances and their combination, before estimating the probability that the first sensor is the correct paired sensor to the set point. We used data from three buildings: A large data set composed of over 12,000 translated data streams for the development and adjustment of the methodology and two smaller data sets for the verification of the transferability of the developed method.The results are strongly dependent on data stream name structure. The results indicate that within the test data set, the pairing sensor to a set point can be automatically limited to three or five sensor candidates, and the algorithm can identify a high number of control loops. For the transfer to the two validation sets, the limitation of sensor candidates is still possible, but identifying pairings and control loops is difficult.