Intelligent steam turbine start-up control based on deep reinforcement learning

The requirement for frequent start-ups and shutdowns is prevalent in turbo-generator units to accommodate fluctuating loads during flexible operations. These cause drastic changes in temperature and stress, leading to instantaneous structural deformations. Hence, research on intelligent start-up control is essential for ensuring safety. In this work, a rotor stress field reconstruction model based on a deep convolutional neural network was first designed. The accuracy of predicting the stress distribution in the critical area reaches 99.7%. The time cost of the trained neural network model is 0.11s in a single case, shortened by 99.8 % with comparison to finite element analysis. Then, a Twin Delayed Deep Deterministic Policy Gradient-based Main Steam Temperature Controller for the Rotor Start-up was proposed and developed. The result shows that the maximum Von Mises stress of the rotor decreases by 14.6 % and 12.2 % in the cold start-up and warm start-up processes with the Controller control. Furthermore, the validity of the Main Steam Temperature Controller was substantiated by comparing its temperature-rising curves with those from the simulated annealing optimization algorithm. The proposed model can effectively increase the start-up speed of the unit and improve the economy while ensuring the safe operation of the unit.