A hybrid reduced-order model of a large-scale generator and power transformer applied in an artificial intelligence-supported power plant control system

Abstract

To optimize the electricity distribution in an electrical grid and integrate power plants with renewable and heat storage energy systems, with a focus on improving energy efficiency while reducing economic costs and emissions, an artificial intelligence method is applied for power plant control and operation monitoring. The effective use of an artificial intelligence method in a power plant can be achieved by implementing real-time digital twins specifically for the most crucial devices, such as power transformers and electric power generators, whose operation and reliability strongly depend on the energy demands and the temperature distribution. However, the development of a power transformer digital twin is based on a complex numerical model that requires high computational demands and large amounts of data for its enhancement. Furthermore, the real-time behaviour of both devices must be considered. Therefore, the main aim of this work is to introduce a hybrid reduced-order model for a large-scale gas-cooled electric power generator and power transformer as the real-time digital twin for a control system. This hybrid approach integrates data gathered from in-field measurements with developed three-dimensional coupled numerical models that can monitor and predict the hot-spot status of both devices at part load, nominal load and overload conditions under different ambient temperatures. The results confirmed the robustness and accuracy of the hybrid reduced-order model within $\pm$8.0 K for all output temperatures due to the accurate predictions of the three-dimensional numerical models within $\pm$5.0 K.