Stress Prediction Processes of Metal Pressure-Bearing Complex Components in Thermal Power Plants Based on Machine Learning

The real-time stress assessment of metal pressure components is one of the key factors in ensuring the safe operation of thermal power plants. To address the challenge of real-time prediction of stress in the key areas of complex special-shaped metal pressure-bearing components in a certain domestic 300 MW thermal power plant, three typical complex metal pressure-bearing components, the main steam pipe tee (MSPT), the steam drum downcomer joint (DDJ) and the header ligament (HL), were taken as research objects. The stress distribution of the three complex metal pressure-bearing components under different conditions was analyzed through the finite element method, and the stress results at the dangerous points were used as samples to establish training sample data. Subsequently, different machine learning methods were employed to train the sample data. The training results indicate that neural networks (NNs) and the Auto-Sklearn Regression (ASR) models can accurately predict the stress of the key parts of complex metal pressure-bearing components in real time. The ASR method demonstrates better performance in stress prediction of the main steam pipe tee, with a prediction accuracy of ≥96%. The NN model shows better prediction for the header ligament, with a prediction accuracy of ≥94%. These research findings provide effective support for the high-temperature lifespan assessment and safe operation of thermal power plants.