Digital twin is a core technology for smart power plants aiming to increase the safety and efficiency of power generation in low-carbon transitions. High-precision modelling of in-service power plant thermal systems plays a key role to develop digital twins, but remains a challenge. There is a lack of high-precision modelling for in-service power plant thermal systems over the full working ranges. This work proposes a hybrid modelling framework combining physical mechanism and operation data to develop grey-box models of thermal systems. Key equipment characteristics are figured out through historical operation data. Moreover, system models consisting of mass and energy balances, process mechanism equations and characteristic equations are implemented. An in-service 660 MW ultra-supercritical double reheat power plant, one of the most advanced thermal power generation technologies, is selected as a case study. The grey-box model of high- and intermediate-pressure thermal system is established. An average simulation error of the model of 0.79% over the full working ranges is achieved. Furthermore, key system characteristics are quantified through the model. It demonstrates the high precision of the proposed modelling method over the full working ranges and provides necessary model support for the digital twin development of thermal power plants.