Domain-informed operation excellence of gas turbine system with machine learning

Abstract

The domain-consistent adoption of artificial intelligence (AI) remains low in thermal power plants due to the black-box nature of AI algorithms and low representation of domain knowledge in conventional data-centric analytics. In this paper, we develop a MAhalanobis Distance-based OPTimization (MAD-OPT) framework that incorporates the AI models and Mahalanobis distance-based constraint to introduce domain knowledge into data-centric analytics. The developed MAD-OPT framework is applied to maximize thermal efficiency and minimize turbine heat rate for an industrial 395 MW capacity gas turbine system. Artificial neural network (ANN) models are trained to predict thermal efficiency, power and turbine heat rate with more than 95% accuracy on the test dataset. The MAD-OPT framework yields 0.1 percentage point relative improvement in thermal efficiency during complete power ramp-up from gas turbine system when operated at ambient temperature of 22 °C than at 34 °C. The robust-optimal values of thermal efficiency (40.99 ± 0.16%) and turbine heat rate (8820 ± 32.82 kJ/kWh) are estimated at the power generation of 305 MW and ambient temperature of 26 °C. More importantly, the framework is successfully deployed to estimate optimal process conditions beyond the design limit of the gas turbine system, addressing the barrier of ANN-based extrapolation in untrained operating space. We also demonstrate that implementing data-centric optimization analytics without incorporating domain-informed constraints may provide ineffective solutions that may not be implementable in the real operation of the gas turbine system. The MAD-OPT framework has demonstrated to adapt with the real-time operating constraints of industrial operations of gas turbine system which may advance the safe and domain-compliant AI adoption to enhance the operation excellence of thermal power systems.