Presented at the 2025 5th International Conference on AI-ML-Systems (AIMLSystems 2025), Bangalore, India, pages 290-298, IEEE.
The paper proposes an adaptive framework for predicting NOx emissions from gas turbines that clusters incoming data with dynamic PCA, based on reconstruction error, and pairs the clusters with SVR, MLP and XGBoost regressors. On real turbine data the clustered approach coped with shifts in operating conditions better than baseline models.
Authors: Kamyar Zeinalipour, Laure Barriere, David Ghelardi, Marco Gori.