Digital twin modeling and fault diagnosis on the feedwater pump system of thermal power plant

The high penetration of renewable energy introduces increased volatility into the power system, presenting more challenging peak shaving and frequency modulation missions for thermal power plants that serve as schedulable power sources within the system. Frequent load cycling place higher demands on the operation stability of core power plant equipment and introduce additional safety risks. As the core system of the power cycle, the operational reliability and energy efficiency of the feedwater pump system (FPS) are crucial. Therefore, this paper focuses on the core equipment and parameters of the FPS, developing a digital twin model based on mechanism modeling method and neural network data-driven modeling method. By comparing the mean absolute error, root mean square error, and coefficient of determination of simulation parameters of different modeling methods, digital twin parameters with higher simulation accuracy were obtained. Utilizing these predicted parameters, an FPS fault diagnosis model (FFDM) for the FPS was developed. By calculating comprehensive deviation degree and the contribution of each parameter, fault early warning and source tracing were achieved, providing guidance for the closed-loop control.