DECOMPOSING BUILDING SYSTEM DATA FOR MODEL VALIDATION AND ANALYSIS USING THE KOOPMAN OPERATOR
Bryan Eisenhower, Tobias Maile, Martin Fischer, Igor Mezić
Large amounts of sensor information is often captured from either realworld building sensors, or virtual building models, for many purposes including control design, fault or aging analysis, and model calibration. Because of the large dimension of this data on both spatial and temporal scales, it is often challenging to come to quick conclu- sions about what information of engineering importance is in the data. In this paper we present an approach to quickly assess spatial information in data based on the spectral content of a certain projection operator. We use operator theoretic methods to capture Koopman modes that represent the spatial content of oscillations in ther- mal quantities. By investigating these modes for different physically significant time-scales (e.g. diurnal, or control system time-scales) we can quickly capture how different parts of a building are responding to load changes at these frequencies (”breathing”). This information helps us to understand anomalies in different aspects of the data, as well as out of phase behavior between zones which may highlight areas of poor control system performance. We present actual and EnergyPlus data from a real building (170K square foot building with approximately 2000 data points) and illustrate how this approach to data analysis and model validation highlights aspects of the data which may otherwise have been overlooked.
- There are currently no refbacks.