David-Fit/Data_Centre_Electrical_Telemetry

Data Centre Electrical Telemetry — Validated Version 1

Dataset Summary

This dataset is a Version 1 validated release of electrical energy telemetry collected from a data centre at Kasetsart University.

The dataset contains telemetry from 24 electrical devices/meters and has been processed and validated using a documented data-processing and engineering-validation pipeline.

The purpose of this release is to provide a validated dataset suitable for engineering analysis and further research involving electrical telemetry, energy monitoring, and data-centre systems.

Original Dataset

This dataset is derived from the original energy telemetry dataset published by:

Chatlatanagulchai, W., & Chantrapornchai, C. (2024). Energy consumption data collection: case study on data center in a Thai University. Energy Informatics, 7, 26.

Original DOI: https://doi.org/10.1186/s42162-024-00327-1

The original dataset repository is:

https://github.com/cchantra/energydata

The original dataset and publication should be cited when using the underlying data.

The original raw telemetry files are not redistributed in this repository. Users requiring the raw source data should obtain them from the original source.

Dataset Contents

The dataset contains telemetry from the following device groups.

Parent and Distribution-Board Meters

• —ADB
• —IUDB
• —NDB
• —UDB1
• —UDB2
• —UDB3

Unit Load Controllers

• —ULC1 through ULC12

Cooling / CRAC Meters

• —CRAC1 through CRAC6

In total, the release contains 24 validated Parquet datasets.

Telemetry Measurements

The telemetry includes electrical measurements such as:

• —Three-phase active power
• —Three-phase apparent power
• —Three-phase capacitive reactive power
• —Three-phase inductive reactive power
• —Phase-level active, apparent, capacitive reactive, and inductive reactive power
• —Phase currents
• —Neutral current
• —Phase-to-neutral voltages
• —Line-to-line voltages
• —Frequency
• —Power factor
• —Cos phi
• —Current THD
• —Voltage THD
• —Maximum demand counters
• —Cumulative energy counters

Processing and Validation

The Version 1 dataset was produced through the following pipeline:

Original CSV/TXT
      |
      v
CSV/TXT → Parquet
      |
      v
Aligned Parquet
      |
      v
Sampling Analysis
      |
      v
Data Audit
      |
      v
Data Cleaning
      |
      v
Cleaned Parquet
      |
      v
Engineering Electrical Validation
      |
      v
Validated Dataset — Version 1

The processing and validation code is available in the associated GitHub repository.

Engineering Validation

The main validation framework uses electrical-engineering relationships and known electrical topology.

Validation includes:

• —Parent-child power conservation
• —Voltage relationships
• —Current relationships
• —Three-phase relationships
• —Neutral current approximation
• —Cumulative energy counter validation
• —Maximum demand checks
• —Physical plausibility checks
• —Power factor / CosPhi consistency
• —Frequency plausibility
• —THD plausibility

Parent-Child Conservation

Parent meters are compared against the summed measurements of their available child meters.

Validated topologies include:

• —UDB1 → ULC1, ULC3, ULC5, ULC7, ULC9, ULC11
• —UDB2 → ULC2, ULC4, ULC6, ULC8, ULC10, ULC12
• —ADB → CRAC1 through CRAC6
• —IUDB → UDB1, UDB2, UDB3

Known topology gaps are documented where available downstream telemetry does not represent the complete parent load. For example, ADB also supplies ALC1 and ALC2, while IUDB also supplies UPS1 and UPS2.

Three-Phase Validation

Three-phase measurements are compared against corresponding L1, L2, and L3 measurements for:

• —Active power
• —Apparent power
• —Capacitive reactive power
• —Inductive reactive power

Confirmed polarity issues identified during engineering validation were corrected for specific measurements, and affected three-phase totals were reconstructed from the corrected phase values.

Voltage Validation

Voltage checks include:

• —Parent-child voltage consistency
• —Phase-to-neutral relationships
• —Line-to-line relationships
• —Approximate √3 relationship between phase and line voltages
• —Phase voltage balance
• —Line voltage balance

Energy Counter Validation

Cumulative energy counters are checked for:

• —Decreasing values
• —Resets
• —Frozen periods
• —Consistency with instantaneous power measurements

Original measured cumulative-energy counters are retained.

Derived cumulative energy counters are provided separately using trapezoidal integration.

Power Factor and CosPhi

Original recorded PowerFactor and CosPhi fields are retained.

Derived PowerFactor and CosPhi measurements are also provided based on active/apparent power relationships.

Version 1 Cleaning and Corrections

The Version 1 processing applies explicitly identified cleaning and correction rules.

These include:

• —Removal of the first four months of telemetry from each dataset
• —Correction of confirmed L2 polarity issues
• —Reconstruction of affected three-phase totals
• —Removal of confirmed invalid negative values for selected device/column combinations
• —Preservation of original measured cumulative-energy counters
• —Addition of derived cumulative energy counters
• —Addition of derived PowerFactor and CosPhi fields
• —Flagging of long frozen cumulative-energy periods without modifying them

The cleaning approach is intentionally conservative.

Data Format

The validated datasets are provided as Parquet files.

Each Parquet file corresponds to one monitored electrical device or meter.

The original raw telemetry remains available from the original dataset source rather than being redistributed in this repository.

Reproducibility

The processing and validation methodology is implemented in the associated GitHub repository.

The complete workflow is:

1. 1.Obtain the original raw telemetry files from the original source.
2. 2.Convert the raw telemetry to aligned Parquet datasets.
3. 3.Perform schema validation.
4. 4.Audit sampling behaviour and timestamp integrity.
5. 5.Perform data-quality auditing.
6. 6.Apply documented cleaning rules.
7. 7.Perform engineering electrical validation.
8. 8.Generate the Version 1 validated datasets.
9. 9.Perform post-clean validation checks.

Code repository: https://github.com/davidfit21/energytelemetryvalidation

Limitations

This is a Version 1 validated dataset and methodology.

Important limitations include:

• —Some parent-child conservation residuals are caused by incomplete downstream metering coverage.
• —Original cumulative energy counters may contain frozen periods or resets.
• —Derived energy counters depend on instantaneous power integration and timestamp continuity.
• —Validation rules depend on the telemetry columns available for each device.
• —Validation also depends on the available electrical topology information.
• —Cleaning corrections are intentionally conservative and are applied only where explicitly identified.
• —This release has not yet been associated with a formal academic publication by the present validation work.

License

The licensing status of the validated dataset is currently not specified.

The underlying telemetry originates from the original dataset and may be subject to separate data-use and redistribution conditions.

Users should consult the original dataset repository and publication and comply with the applicable terms before redistributing or otherwise using the underlying data.

Citation

When using the validated Version 1 dataset, please cite this release:

Fit, D. (2026). Data Centre Electrical Telemetry — Validated Version 1. Zenodo. https://doi.org/10.5281/zenodo.22143264

The underlying original dataset should also be cited:

Chatlatanagulchai, W., & Chantrapornchai, C. (2024). Energy consumption data collection: case study on data center in a Thai University. Energy Informatics, 7, 26. https://doi.org/10.1186/s42162-024-00327-1

Related Resources

Original dataset repository: https://github.com/cchantra/energydata

Original publication: https://doi.org/10.1186/s42162-024-00327-1

Processing and validation code: https://github.com/davidfit21/energytelemetryvalidation

Zenodo — Version 1 archival release: https://doi.org/10.5281/zenodo.22143264