28 months
Industrial field validation period for the reference PEMS case.
PEMS evidence note
What the 28-month PEMS field validation shows, and what it does not show.
A Predictive Emissions Monitoring System (PEMS) can be technically evaluated over long industrial operating periods when same-period process data and emissions data are available. The reference case gives Zolena Lab a disciplined evidence base for NOx PEMS pilots, data readiness review, and model performance review.
Why this matters
Long-duration field validation is different from a short model demonstration.
The reference PEMS case reported 28 months of industrial field validation for a NOx predictive monitoring application. That duration matters because industrial units experience changing load, operating modes, sensor conditions, maintenance events, and data-quality issues over time.
For Zolena Lab, the case is not used as a universal sales promise. It is used as a technical precedent for asking the right questions: whether the customer's data are complete enough, whether timestamps align, whether operating ranges are represented, and whether model performance should be assessed under different operating conditions.
Reference metrics
Key numbers from the reference case.
99.93%
Reported data availability. This is not prediction accuracy.
MAE 0.5982
Reported model error metric for the specific reference case.
r = 0.9451
Reported correlation metric for the specific reference case.
The reference case also reported a 0.14% total-emissions difference in the test set. These values describe one reference case and should not be read as guaranteed results for other facilities.
What the evidence supports
How Zolena Lab uses the field-validation evidence.
One-unit pilot design
The evidence supports starting with one clearly bounded unit, same-period process and NOx data, held-out validation, and a written technical report.
Data readiness review
The evidence shows why tag naming, units, sampling frequency, timestamp alignment, quality flags, and commissioning data preservation matter before a model is built.
Second opinion work
The evidence helps frame technical questions about validation logic, data sufficiency, error, bias, operating-condition coverage, and model-versus-instrument divergence.
What it does not prove
Technical evidence must be kept inside its boundary.
Not universal performance
The 28-month result does not guarantee that every turbine, boiler, CHP unit, HRSG, or industrial facility will achieve the same metrics.
Not automatic compliance approval
Technical validation is not regulatory approval, RATA, certification, legal assurance, or automatic permission to use model estimates for official reporting.
Not CEMS replacement by default
Any CEMS substitution or official reporting use must be separately confirmed under local rules, permit conditions, facility approval, and qualified third-party requirements.
Academic source
The evidence traces back to published industrial PEMS research.
Reference: Si, M., Tarnoczi, T. J., Wiens, B. M., & Du, K. (2019). Development of Predictive Emissions Monitoring System Using Open Source Machine Learning Library - Keras: A Case Study on a Cogeneration Unit. IEEE Access. DOI: 10.1109/ACCESS.2019.2930555.
Apply this evidence carefully
Use the reference case to shape a scoped pilot, not to assume a guaranteed facility result.
Start with unit type, data coverage period, NOx measurement context, and a preliminary tag list.