摘要
为了预测船舶柴油机NOx排放特性,从初始权值的选取及学习率动态优化对BP算法进行了改进,通过改进的均匀试验设计法,对少量具有代表性、易于测试的工况进行NOx排放测试,利用BP神经网络建立了船舶柴油机NOx排放特性预测模型并进行了计算,与实测的4种工况进行比较.结果表明,第1工况的NOx排放浓度相对误差为3.7%,NOx比排放的相对误差为4.3%,而其他各工况的NOx排放浓度相对误差在2.4%以内,NOx比排放相对误差在2.9%以内.因此,该模型预测精度较高,与试验结果吻合良好,能有效地预测船舶柴油机NOx排放特性.
In order to predict marine diesel engine's NOx emission characteristics,the BP algorithm is improved by selecting initial weights and using dynamic optimal learning rate,NOx emission samples are got from the bed test on a marine medium diesel engine by a new method-variable edge uniform distribution and experimental design(U-D design),and an NOx emission prediction model based on BP neural network is set up.Compared with four test modes,simulation results show that there are 3.7% NOx emission relative error and 4.3% specific NOx emission relative error in the first test mode,and there are less than 2.4% NOx emission relative error and less than 2.9% specific NOx emission relative error in the other test modes.Therefore,the NOx emission prediction simulation well coincides with test modes,the BP neural network is an effective way to predict marine diesel engine's NOx emission characteristics.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2007年第1期6-10,共5页
Journal of Harbin Engineering University
基金
国家自然科学基金资助项目(50276006)
