摘要
传统的NOx排放模型都是基于Zeldovich链式反应,其大量的计算无法满足HILSS(硬件在环仿真系统)实时仿真的要求。而根据各种影响因素与NOx生成量之间的映射关系,用神经网络方法来构建NOx排放模型是一种同时兼顾实时性和准确性的解决方案。所建立的基于BPNN的NOx排放模型,采用贝叶斯正则化训练算法提高BP网络的推广能力,具有简单、可靠和通用的特点,可以在一定程度上预测发动机瞬态工况的NOx排放。
The traditional NOx emission model based on Zeldovich chain reaction needs a lot of calculation time, couldn＇t meet the real-time demand of HILSS. So the neural network NOx emission model base on the reflection relationship between the amount of NOx and some direct influence factors is a good solution to the contrast of accuracy and real time demand. The BPNN is trained by Bayesian regularization, which updates the weight and bias values according to Levenberg-Marquardt optimization and so as to produce a network that generalizes well. The verify of NOx emission NN model shows it is simple, reliable and universal to other type of diesel engine, and this model even could predict the state of transient engine NOx emissions accurately to some extent.
出处
《车用发动机》
北大核心
2007年第2期37-39,共3页
Vehicle Engine
基金
高等学校博士学科点专项科研基金资助项目(20040487038)
关键词
柴油机
氮氧化物
排放控制
神经网络
实时仿真
diesel engine NOx emission eortrol
neural network
real-time simulation
