Using Machine Learning Methods to Predict the Ductile-to-Brittle Transition Temperature Shift in RPV Steel Under Different Pulse Current Parameters

The reactor pressure vessel (RPV) is susceptible to brittle fracture due to the influence of ion irradiation and high temperature, which presents a significant risk to the safe operation of nuclear reactors. It has been demonstrated that pulsed electric current can effectively address the issue of embrittlement in RPV steel. However, the relationship between pulse parameters (duty ratio, frequency, current, and time) and the effectiveness of pulse current processing has not been systematically studied. The application of machine learning methods enables autonomous exploration and learning of the relationship between data. Consequently, this study proposes a machine learning method based on the random forest model to establish the relationship between the parameters of electrical pulses and the repair effect of RPV steel. A generative adversarial network is employed to enhance data diversity and scalability, while a particle swarm optimization algorithm is utilized to optimize the initialization weights and biases of the random forest model, aiming to improve the model’s fitting ability and training performance. The results indicate that the coefficient of determination R-square (R²), root mean squared error and mean absolute error values are 0.934, 0.045, and 0.036, respectively, suggesting that the model has the potential to predict the performance recovery of RPV steel after pulsed electric field treatment. The prediction of the impact of pulse current parameters on the repair effect will help to enhance and optimize the repair process, thereby providing a scientific basis for pulse current repair processing.