Physics-constrained online monitoring for continuous heat-transfer performance identification in flexibly operated thermal power plants

The increasing demand for peak shaving and load following in coal-fired power plants requires reliable online monitoring of heat-transfer equipment under dynamic operating conditions. Conventional steady-state methods apply only to quasi-steady segments and fail during dynamic operation. Response delays create a temporal mismatch between boundary measurements and the actual system state, while small terminal temperature differences amplify measurement noise. To address these issues, this study proposes a physics-constrained residual decomposition method for continuous heat-transfer performance identification. A calibrated distributed physical model provides an instantaneous steady-state reference under current boundary conditions. The method decomposes the difference between measured and predicted outlet temperatures into delay bias, equipment variation, and measurement noise. Removing the delay component reconstructs a pseudo-steady state consistent with the current operating condition. Validation on the condenser system of a large thermal power unit shows that the approach reduces the root-mean-square error of heat-transfer coefficient identification by 68.6% during pump switching and 37.8% during load ramping, and decreases the energy conservation residual mean-square error by 61.2%. Monitoring coverage expands from quasi-steady segments accounting for only 18.9–35.1% of operating time to the full operating range, enabling continuous tracking of fouling-induced performance degradation without additional sensors.