Benchmarking Data-Driven Time Series Classification Methods for a real-world Building HVAC System: Dataset

Original Publication: Michael Hadwiger, Thomas Hirsch, Samina Kadkhoda Masoumali, and Gerald Schweiger. 2026. Benchmarking Data-Driven Time Series Classification Methods for a real-world Building HVAC System. In The 13th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys ’26), June 22–25, 2026 Banff, AB, Canada. ACM, New York, NY, USA, 11 pages. https://doi.org/10.1145/3744256.3812577 Acknowledgment: The research leading to these results was conducted within the project Self2B (project number 920143), funded by the Austrian Research Promotion Agency (FFG). Dataset description: This dataset accompanies the paper "Benchmarking Data-Driven Time Series Classification Methods for a real-world Building HVAC System" (BuildSys '26). It contains anonymised sensor time-series data from a mixed-use office and laboratory complex comprising 30 buildings, 1832 sensors, and 18 sensor classes, spanning over 23 months of continuous measurements. Dataset contents: sensor_building_mapping.csv — maps each anonymised sensor ID to its building and sensor class data_anon/ — raw time-series data, one Parquet file per sensor (timestamp + value, irregularly spaced) features_Prab2024/, features_Hong2015/ — pre-computed feature vectors for two baseline feature sets from related literature features_TA/, features_TSA/, features_TSAR/ — pre-computed feature vectors for the novel time-aware (TA) and time- and slope-aware (TSA) feature sets introduced in this work, including a resampled variant (TSAR) features_catch24/ — pre-computed feature vectors using the catch22/catch24 canonical time-series feature set data_LLM/ — sensor segments and results for the LLM-based classification pipeline All feature folders are split by building (B01–B30) and weekly time window and include per-class summary statistics and data removal logs. Sensor classes: The dataset covers 18 sensor classes at subsystem level (Level 4) and 6 classes at system level (Level 3), spanning temperature, humidity, and differential pressure sensors across ventilation, heating, cooling, and domestic hot water systems: HuVentEx, HuVentSu, PrVentEx, PrVentRo, PrVentSu, TeCoolOA, TeCoolRe, TeCoolSt, TeCoolSu, TeDoHWPi, TeDoHWSt, TeHeatRe, TeHeatSu, TeVentEx, TeVentHW, TeVentOA, TeVentRo, TeVentSu. Usage: Download the archive and extract its contents into the data/ folder of the accompanying code repository. Pre-computed features can be used directly; raw time-series can be reprocessed using the provided scripts. The source code is available on GitHub and archived on Zenodo (see related identifiers).