基于Elman神经网络的超超临界机组汽水分离器应力在线软测量模型

摘要

根据超超临界锅炉汽水分离器的结构特点建立了三维有限元模型，以某1000MW机组为例模拟了其启动过程的总应力场．将有限元法和神经网络法相结合，以有限元计算结果作为训练样本，以介质压力和筒体壁温序列为辅助变量，建立了基于Elman神经网络的分离器应力动态软测量模型，通过模型的训练，确定了准确的应力预测模型结构．应用电厂实际运行监测数据对所建立的Elman网络软测量模型进行验证，结果表明：模型计算结果可很好地逼近有限元结果，预测精度高，实时性好，可为锅炉寿命的在线监测提供数据支持．%Based on structural features of steam-water separator for ultra supercritical boilers, a 3D finite element model was established for thermal analysis purpose, with which the total stress field was simulated during start-up of a 1 000 MW unit. Combining the finite element method with neural network algorithm, a dynamic stress soft-sensing model was set up based on Elman neural network, by taking both the medium pressure and wall temperature series as auxiliary variables, and the finite element calculation results as training samples, of which the structure was subsequently determined after model training. The soft-sensing model was finally verified with actual operating and monitoring data of the power unit. Results show that the model calculation results can well approach that of finite element analysis. Featured by high precision and good real-time performance, the model may serve as a reference for on-line life monitoring of power plant boilers.