Real-Time Compression of Time Series Building Performance Data

AbstractIf building performance simulations are to fully benefit from increasing quantities of sensor data, subsets of large datasets must be efficiently extracted at varying levels of detail. A key issue with time series data is that relevant time scales vary by orders of magnitude depending on the desired analysis. To ensure that a subset of a time series is available when needed at an appropriate resolution, lossy compression can be applied in real time as data is acquired. We propose an algorithm that compresses a piecewise constant time series by merging segments within a sliding time window. This procedure tends to preserve prominent edges and spikes. While building control system dashboards and simulation tools often average data over fixed time periods (e.g. hourly averages), the proposed method achieves lower errors for the same compression ratio and provides better support for signal processing, data visualization, and simulation.