A 30-Day Raw Telemetry Dataset Package for an Indoor Ice Rink Refrigeration System

# 30-Day Indoor Ice Rink Telemetry Dataset Package ## 1. Package contents - `raw_30d_telemetry.csv` — raw row-level telemetry archive extracted from ClickHouse. - `minute_30d_from_raw.csv` — minute-aligned intermediate table obtained from the raw archive. - `data_30d_ml_from_raw.csv` — cleaned minute-level analytical table. - `preprocess_ice_rink_30d_raw_telemetry.py` — preprocessing script that reconstructs the minute-level files from the raw archive. - `data_dictionary.csv` — machine-readable description of the released files and columns. - `signals_mapping.csv` — mapping from raw signals to cleaned variables. - `requirements.txt` — minimal Python dependencies required to run the preprocessing script. - `LICENSE.txt` — license notice for the supplementary dataset package. ## 2. Observation window The released package covers the observation window from 2025-03-03 00:00:00 to 2025-04-02 00:00:00 in the Europe/Moscow timezone. The raw archive preserves timestamps with an explicit `+03:00` offset. The intermediate and cleaned minute-level tables use naive minute timestamps consistent with the internal `data.csv` workflow. ## 3. File descriptions - `raw_30d_telemetry.csv` — raw row-level archive restricted to the twenty selected telemetry signals. - `minute_30d_from_raw.csv` — minute-level intermediate table produced by rounding timestamps down to the nearest minute, averaging repeated observations within each minute, pivoting to a wide table, and reconstructing the complete minute calendar. - `data_30d_ml_from_raw.csv` — cleaned minute-level analytical table produced by continuity restoration and range-based plausibility filtering. ## 4. Raw schema The raw archive contains the following columns: - `timestamp` - `device_name` - `measure_name` - `channel` - `phase` - `signal_name` - `value` ## 5. Processing steps 1. Filter the source archive to the twenty signals used in the refrigeration dataset. 2. Round timestamps down to the nearest minute. 3. Average repeated observations within each minute for the same signal. 4. Pivot the signal stream to a wide minute-level table. 5. Aggregate composite power variables for compressors and heating. 6. Restore continuity by forward filling and remove the unresolved leading segment. 7. Apply plausibility filtering for the supervisory setpoint, compressor power, and heating power. ## 6. How to reproduce 1. Install the dependencies listed in `requirements.txt`. 2. Run: ```bash python preprocess_ice_rink_30d_raw_telemetry.py ``` 3. The script recreates `minute_30d_from_raw.csv` and `data_30d_ml_from_raw.csv` from `raw_30d_telemetry.csv`. ## 7. Notes and limitations - The raw file may contain repeated observations within the same minute for the same signal. - The intermediate minute-level table retains missing values and should not be treated as a final analytical dataset. - The cleaned minute-level table is unlabeled and contains no engineered features. - The release is limited to a 30-day window and to twenty selected telemetry signals. ## 8. License and citation The supplementary dataset package is distributed under the CC BY 4.0 license. When reusing the package, cite both the dataset paper and the supplementary dataset package submitted with it.