基于RBF-BP混合神经网络的烧结烟气NOx预测

摘要：对烧结烟气NOx生成量进行预测，能为烧结NOx源头和过程减排提供有效指导。利用BP神经网络模型和RBF神经网络模型对烧结烟气NOx进行了预测，在此基础上结合BP模型自适应学习能力强和RBF模型快速收敛的特性，采用优化模型结构、设立连接层的方法，构建RBF BP混合神经网络模型进行了NOx预测研究，并对3种模型的预测结果进行了对比分析。研究表明，3种神经网络模型中，RBF-BP混合模型的均方根误差为11.37mg/m3，平均绝对误差为7.14mg/m3，最大绝对误差为35.47mg/m3，最小绝对误差为0.0083mg/m3，各评价指标均为3种模型中最优，混合神经网络模型的预测数据稳定性更好，结果拟合程度更高且收敛速度最快。采用混合模型预测NOx能有效消除烟气NOx生成量反馈延迟。

NOx prediction of sintering flue gas based on RBF-BP hybrid neural network

Prediction of NOx production in sintering flue gas can provide effective guidance for the emission reduction of sintering NOx source and process. BP neural network model and RBF neural network model were used to predict the sintering flue gas NOx. On the basis of analyzing and studying two kinds of neural network models, combining the strong adaptive learning ability characteristics of BP model and the fast convergence characteristics of RBF model, using the method of optimizing model structure and setting up connection layer, the RBF BP hybrid neural network model is constructed for NOx prediction, and the prediction results of the three models are compared and analyzed. The results show that the RMSE of RBF-BP hybrid model is 11.37mg/m3, the MAE is 7.14mg/m3, the maximum absolute error is 35.47mg/m3, and theminimum absolute error is 0.0083mg/m3, each evaluation index is the best among the three models. Compared with the other two single neural network models, the hybrid neural network model has better stability of prediction data, higher fitting degree of results and the fastest convergence speed. Predicting NOx by using mixed model can effectively eliminate the feedback delay of flue gas NOx production.