Data provenance: This model is trained and validated on the UCI "Gas Turbine CO and NOx Emission Data Set" — real hourly measurements from an actual gas turbine plant in Turkey (2011–2015), published by Kaya et al. (2019). No public Kenyan equivalent dataset exists. KenGen is used as the example deployment target only. Actual deployment at a KenGen facility would require KenGen's own operational historian data.
"Trained and validated on real published gas turbine data (Turkey, UCI repository) to prove the method; no equivalent public Kenyan dataset exists yet — next step is partnering with a local plant for their own data."
"Trained and validated on real published gas turbine data (Turkey, UCI repository) to prove the method; no equivalent public Kenyan dataset exists yet — next step is partnering with a local plant for their own data."
Module B — Predictive PEMS: Carbon Monoxide (CO)
XGBRegressor trained on 3 years, evaluated on 2-year test set.
R² (CO)
0.6037
RMSE (CO)
1.6803 mg/Nm³
Module B — Predictive PEMS: Nitrogen Oxides (NOX)
R² (NOX)
-0.1717
RMSE (NOX)
12.2235 mg/Nm³
Compliance — CO (latest predicted, example limit)
Within Limit
Compared against EU IED 2010/75 example limit for new gas turbines (100 mg/Nm³); Kenya-specific regulations pending.
7.68
mg/Nm³
Limit: 100 mg/Nm³
Compliance — NOX (latest predicted, example limit)
Exceedance Detected
Compared against EU IED 2010/75 example limit for new gas turbines (50 mg/Nm³); Kenya-specific regulations pending.
100.90
mg/Nm³
Limit: 50 mg/Nm³