Modeling for Measurement Process of Thermal Resistor Tube Socket in Primary Loop Hot Leg of HPR1000 Using Machine Learning

Current assessments of primary loop hot leg temperature in nuclear power plants rely on conservative uncertainty quantification, where underexplored process measurement uncertainties hinder operational optimization. To address this, we develop a model of the thermal resistor tube socket temperature measurement process in HPR1000 using a convolutional neural network modeling method with a loss function with physical information and a dynamic time factor (DTF-CNN-PINN). Validation against Fuqing Unit 5 operational data demonstrates: (1) high-fidelity temporal coherence under dynamic thermal-hydraulic conditions; (2) a mean squared error of 0.0076 in temperature field reconstruction; (3) a 2°C accuracy improvement over computational fluid dynamics (CFD) benchmarks. This work systematically models the temperature measurement process in nuclear reactor primary pipelines, while also providing analytical redundancy for physical sensors.