用于HCCI发动机燃烧状态参数辨识的神经网络  

Neural Networks for Observing Gasoline HCCI Combustion Phasing

作　　者：孙艳辉[1] 谢辉[2] 夏超英[1] 

机构地区：[1]天津大学自动化学院,天津300072 [2]天津大学内燃机燃烧学国家重点实验室

出　　处：《小型内燃机与摩托车》2007年第6期25-29,共5页Small Internal Combustion Engine and Motorcycle

摘　　要：本文分别利用Elman网络、BP网络和RBF网络从离子电流信号辨识HCCI发动机的燃烧相位CA50,并对三种模型的各项性能进行了比较。该方法首先提取每个循环的离子流信号的4个特征信息,用提取的特征信息和发动机转速以及4个控制参数经归一化处理后,输入给神经网络,计算出CA50。本研究以基于全可变气门机构的汽油HCCI发动机为对象,选取了台架试验中6个典型的HCCI动态变负荷过程数据作为训练样本,另两个动态变负荷数据为测试样本。测试结果表明:Elman网络的训练耗时明显最长,计算时间稍长于BP网络和RBF网络;RBF网络具有最好的拟合精度,但泛化能力最差,而Elman网络的泛化能力最好,Elman观测器具有更强的抗干扰性。综合考虑Elman网络更适合于HCCI发动机燃烧状态参数辨识。Elman network, BP network and RBF network are used to detect combustion phase CA50 of HCCI engine from ion current signal in this paper, and comparison are made among the performances of the three models. At first, four characteristic signals such as the position of peak, the start point, the end point and the inflection point, are extracted from the ion current signal of every cycle. Then the 4 characteristic signals, en- gine speed together with 4 control parameters are normalized. At last, combustion phase CA50 is calculated by neural network with the 9 normalized parameters above as its input. The study is based on HCCI gasoline en- gine with a fully variable valve actuating system, and 6 sets of typical dynamic data based on changing load are used as training set, the other two dynamic data are used as the test sets. The results demonstrate that Elman network takes longest time to finish training, and its calculation time is a bit longer than the other two net- works. RBF network gets the best fitting precision, but the worst generalization ability, and Elman network works with the best generalization ability. Compared with BP network and RBF network, Elman network shows great advantages in resistance to disturbance. On the whole, Elman network fits for observing HCCI engine combustion phasing much better.

关 键 词：HCCI汽油机 燃烧相位观测 离子电流 ELMAN网络 BP网络 RBF网络 

分 类 号：TK401[动力工程及工程热物理—动力机械及工程]