This study investigates the performance improvement of a model to predict gas turbine maximum power output through the application of data augmentation techniques. The accurate prediction of a gas turbine maximum output is crucial for stable power system operation and economic dispatch, as it is complexly determined by environmental conditions (such as outside temperature, humidity, and pressure) and equipment status. A prediction model was initially developed by integrating gas turbine operating data and corresponding meteorological data from a combined cycle power plant. A critical challenge in this process was the limited volume and variability of the real-world dataset, which constrains the model's accuracy and generalizability. To overcome this limitation, we introduced and applied data augmentation techniques to generate a more robust and diverse training set. The prediction model, utilizing the XGBoost and LightGBM algorithms, showed significant performance enhancement after applying data augmentation. Specifically, the application of data augmentation successfully reduced the RMSE to approximately 2 MW, achieving a maximum error reduction of 18% compared to the model trained on the original data alone. These results empirically demonstrate the high effectiveness of data augmentation techniques in mitigating data scarcity issues and substantially improving the prediction accuracy and reliability of gas turbine maximum output models.
Kim et al. (Mon,) studied this question.