Data-Driven Surrogate Modelling of Modified Gas Turbine Cycles Using Machine Learning for Real-Time Performance Prediction

Gas turbine power plants remain a key component of flexible electricity generation, yet their performance is strongly influenced by ambient conditions, operating parameters, and retrofit configurations. Conventional thermodynamic models provide high accuracy but are computationally intensive and unsuitable for real-time applications, particularly for modified gas turbine cycles with increased system complexity. This study presents a data-driven surrogate modelling framework for real-time performance prediction of conventional and modified gas turbine cycles by integrating first-principles thermodynamic simulation with advanced machine learning techniques. A conventional gas turbine and three modified configurations incorporating inlet air cooling, regeneration, heat-recovery steam generation, steam injection, dual combustion, and staged turbine expansion were modelled in ASPEN HYSYS to generate a large parametric dataset across realistic operating ranges. Artificial neural network (ANN), random forest (RF), and extreme gradient boosting (XGBoost) models were developed to predict key performance indicators, including thermal efficiency, net power output, and emission intensity, and were benchmarked against multiple linear regression models. The thermodynamic results show that all modified configurations significantly outperform the conventional cycle. The optimal configuration achieved a thermal efficiency of approximately 47%, an increase in power output exceeding 40%, and a substantial reduction in emission intensity. The machine learning surrogate models demonstrated excellent predictive accuracy, with coefficients of determination greater than 0.95 for all outputs, and provided several orders-of-magnitude reductions in computational time compared with detailed process simulations. Feature-importance analysis identified turbine inlet temperature, compressor pressure ratio, and fuel mass flow rate as the dominant performance drivers, consistent with thermodynamic principles. The proposed framework enables rapid and accurate performance prediction and provides a practical foundation for deploying digital twins in gas turbine power plants. It offers a scalable pathway for enhancing the efficiency, flexibility, and environmental performance of existing gas turbine-based power generation systems. Keywords: Gas turbine; Machine learning; Surrogate model; Digital twin; ASPEN HYSYS; ANN; XGBoost; Random Forest, Performance prediction.