One-minute records spanning 18 March 2022 to 28 February 2023.
What the dataset adds
The associated study reports this count after preparation. It is not the raw-file row count.
Temperatures, flows, digester levels, total biogas and burner/boiler distribution.
Biogas channel is total biogas flow in CFM. It is not methane. GFIS does not relabel it or merge it with the DS-05 measured-methane target.Two complementary evidence streams
DS-05 supplies 138 controlled reactor records and 86 measured methane targets for chronological persistence, XGBoost and LSTM comparison. DS-06 supplies plant-scale operational density. Their targets and cadences remain separate.
Data engineering that can be defended
Immutable source
The 85 MB raw CSV remains unchanged. SHA-256: 8d56ac54…04440.
Order made explicit
The physical CSV row order is not chronological. Replay orders timestamps before applying any stride.
All rows queryable
A 500,400-row SSD SQLite index provides pagination without sending the full file to the browser.
Quality audit
Zero invalid timestamps, zero duplicate timestamps and 19 non-one-minute discontinuities.
Plant-twin contract
Temperature, feed and gas channels can drive observed replay; simulated states stay separately labelled.
Next benchmark
Reproduce the published cleaning procedure before chronological total-biogas model training.
The digital twin connection
The 3D plant is the visual state layer. An evidence adapter maps measured SCADA channels to equipment state, a Level 2 engine generates scenario trajectories, and experiment memory stores the source, timestamp, configuration and output. Observed and simulated modes are visibly distinct.


Open interactive 3D plant Open separate gasification prototype