[Objective] In the context of the carbon neutrality objective and the ongoing global energy transition

the installed capacity of renewable energy power generation such as wind and solar energy is continuously increasing. Due to the unstable output of renewable energy sources

the role of coal-fired power plants in ensuring renewable energy accommodation and grid stability has become increasingly clear. [Methods] Establishing an accurate and efficient dynamic model is the key to improving the operational flexibility of coal-fired power plants

as it lays a solid foundation for the operation optimization and control strategies. For this reason

this paper proposes a novel dynamic modeling method for the boiler-turbine coordination system of an ultra-supercritical coal-fired power plant based on the physics-informed neural network(PINN)

with the core purpose of accurately predicting key operational parameters including main steam pressure

separator steam enthalpy and power generation. The proposed PINN takes the conventional feedforward neural network(FNN) as the basic network framework. Meanwhile

the basic dynamic physical laws of energy-mass conversion in the boiler-turbine coordination system are identified and embedded into the learning framework of the neural network in the form of loss functions. Such a design integrates the physical mechanisms of the plant with data-driven learning

thus enabling the network to effectively learn complex and highly non-linear data features while strictly satisfying basic physical laws of the boiler-turbine system. [Results] The verification results indicate that the proposed PINN model achieves improved overall accuracy

inherent interpretability and strong generalization ability. In data-sufficient scenarios

the coefficient of determination(R2) indexes of both one-step and multi-step predictions generated by the PINN method are both higher than 0.99

demonstrating excellent and robust fitting performance between the predicted values and the actual measured data. In data-scarce scenarios

which is common in practical engineering applications

the PINN method still maintains high prediction accuracy. The root mean square error(RMSE) index for multi-step prediction of main steam pressure is less than 1%

and the RMSE indexes for multi-step predictions of separator steam enthalpy and power generation are both less than 0.75%. In both scenarios

the prediction performance of the proposed PINN method is significantly superior to that of the traditional neural network

which further verifies the effectiveness and superiority of the proposed modeling method. [Conclusion] This study provides an effective dynamic modeling idea for the boiler-turbine coordinated system

which is of great significance for improving the control performance and flexibility of ultra-supercritical coal-fired power plants in engineering applications.