基于具有长短时记忆功能的深度循环神经网络(Deep Recurrent neural network with Long-short term memory,DRNN-LSTM),利用某厂650WM燃煤锅炉一个月的历史运行数据,建立了SCR烟气脱硝系统出口NOx排放预测模型。DRNN-LSTM网络同一隐藏层的所有循环体中的参数矩阵相同,需要学习的参数个数少,训练模型具有很高的稳定性。测试结果与定量分析表明,DRNN-LSTM模型计算时间与其他传统数据模型相近,但拟合效果与预测精度明显高于其他模型,说明该模型收敛速度快、泛化能力强,可准确描述SCR脱硝系统的反应过程,能够有效应用于电厂烟气脱硝系统出口NOx浓度的预测中。

Based on the Deep Recurrent neural network with Long-short term memory (DRNN-LSTM) and one-month historical operation data of a 650WM coal-fired boiler in a plant, established the NOx emission prediction model of SCR flue gas denitrification system. The parameter matrix in all loops of the same hidden layer of the DRNN-LSTM network is same, the number of parameters to be learned is small, and the training model has high stability. The test results and quantitative analysis show that the calculation time of DRNN-LSTM model is similar to other traditional data models, but the fitting effect and prediction accuracy are significantly higher than other models. It shows that the model has fast convergence speed and strong generalization ability, can accurately describe the reaction process of SCR denitrification system. It can be effectively applied to the prediction of export NOx concentration.

TK39

上海市科委地方能力建设项目（18020500900）；上海市自然科学基金（19ZR1420700）