基于回归神经网络的柴油机NOx排放预测及影响分析  

NOx emission prediction and impact analysis of diesel based on recurrent neural network

作　　者：伦智梅 张振京 宋业栋 谢熙 顾昕岑 LUN Zhimei;ZHANG Zhenjing;SONG Yedong;XIE Xi;GU Xincen(State Key Laboratory of Engine Reliability,Weifang 261061,China;Weichai Power Company Limited,Weifang 261061,China)

机构地区：[1]内燃机可靠性国家重点实验室,山东潍坊261061 [2]潍柴动力股份有限公司,山东潍坊261061

出　　处：《山东理工大学学报（自然科学版）》2021年第3期81-86,共6页Journal of Shandong University of Technology:Natural Science Edition

摘　　要：针对柴油发动机NO x排放量的预测问题与其影响参数进行研究。通过构建训练神经网络回归算法模型(recurrent neural network,RNN),基于发动机试验过程中采集的大量样本数据训练和预测NOx排放情况;预测结果表明,该网络模型的预测值与目标输出值之间的误差满足实际工程需求,具有良好的预测精度和泛化性能。基于参数最优的神经网络回归模型,采用平均影响值(mean impact value,MIV)算法定量地分析各输入参数对目标输出参数的影响权重,进一步分析影响NOx排放的关键影响因素;结果表明,本文采用的MIV算法可以准确判断各输入因子对网络输出的相对影响权重,实现关键影响因子的分析。The estimation on NOx emission of diesel engine and its influence factor were studied.Based on the large amount of sample data collected from the diesel engine tests,the recurrent neural network(RNN)was constructed and trained to predict the diesel′s NOx emission load.The test results show that the error between the predicted value of the RNN model and the target output value meets the actual engineering requirements and has a good prediction accuracy.Based on the constructed RNN model,the mean impact value(MIV)algorithm was used to quantitatively analyze the impact weight of each input parameter on the target output parameter,so as to analyze the key influencing factors affecting NOx emissions.The result shows that the MIV algorithm adopted in this paper can accurately judge the relative impact weight of each input parameter on the model′s output and realize the analysis of key influence factors.

关 键 词：NOX排放 回归神经网络 平均影响因子 预测 

分 类 号：TK421[动力工程及工程热物理—动力机械及工程] TP183[自动化与计算机技术—控制理论与控制工程]