Summary
Forecast errors in power generation lead to significant imbalance penalties, reducing the economic efficiency of plant operations. In competitive electricity markets, deviations from scheduled generation are penalized through a dual-price mechanism, and inaccurate predictions can substantially diminish plant profit margins. This research systematically compares the performance of multiple regression techniques for predicting the net power output of a 165 MW triple-flash and binary cycle geothermal power plant (GPP) in Western Anatolia. The primary aim is to statistically quantify the effects of key operating parameters—inlet temperature, mass flow rate of geothermal fluid, and ambient air temperature—on net electrical power output. The analysis utilises 365 daily averaged observations. Comprehensive data pre-processing, including outlier elimination, feature selection, and normalization, was conducted to ensure statistical reliability prior to model training. Five regression methods were developed: Multiple
Read more Read lessLinear Regression (MLR), Polynomial Regression (PR), Decision Tree Regression (DTR),
Random Forest Regression (RFR), and Gradient Boosting Regression (GBR). To prevent overfitting and enhance robustness, 10-fold cross-validation and targeted sensitivity analysis of structural hyperparameters were implemented. Model performance was evaluated using R²,
Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). The GBR model, optimized through sensitivity analysis, achieved the highest predictive accuracy with a test R² of 0.762 and the lowest MAE of 1.637 MW, representing a 35.3% reduction in prediction error compared to the baseline linear model. Crucially, the achieved MAE remains well within the 5% regulatory tolerance margin, effectively mitigating the risk of imbalance penalties. These findings establish GBR as the most effective approach for minimizing forecast uncertainty in this context.
Additional informations
| Publication type | Session Materials |
|---|---|
| Reference | D2_11682_2026 |
| Publication year | |
| Publisher | CIGRE |
| Country | Turkiye |
| Study committees | |
| File size | 1 MB |
| Price for non member | 30 € |
| Price for member | 30 € |
Authors
MERCAN Fatih - Turkish Electricity Transmission Corporation; YILDIZ Ahmet - Afyon Kocatepe University; YILMAZ Ceyhun - Sakarya University
Keywords
Geothermal Power Plant, Net Power Forecasting, Gradient Boosting Regressor, Ensemble Learning, Hyperparameter Optimization
