利用退役转子及其对应的实际历史运行工况，开展了基于NARX神经网络的汽轮机转子关键位置温度发展趋势实时预测的研究。利用有限元方法，计算了历史运行工况下转子的高保真温度场，获取了关键位置的温度变化趋势。采用NARX神经网络，构建了转子进口蒸汽温度和功率与转子关键位置温度之间的数学关系，通过将输出参数闭环到输入参数集，形成了闭环神经网络结构，实现实时多步预测。预测结果表明，该模型对温度变化趋势的实时预测效果较好，且可以快速校正误差较大的预测结果。

An exservice rotor and its corresponding historical operating conditions are used to carry out a research on the realtime prediction of the temperature evolution of the steam turbine rotor based on the NARX neural network.Firstly,the highfidelity temperature field of the rotor under historical operating conditions is calculated by FEM,and the temperature trend at a key position is obtained.Furthermore,NARX neural network is established to construct the mathematical relationship between the rotor inlet steam temperature and power and the temperature at the key position of the rotor.By looping the output parameters to the input parameter set,a closedloop neural network structure is formed,realizing the realtime multistep prediction.The prediction results show that the model has a good realtime prediction effect on the temperature change trend,and can quickly correct the prediction results with large errors.

TK267