研究神经网络理论用于内燃机排放预测模型时学习样本的选取方法,针对内燃机工况变化的特点,对传统的正交设计法进行了改进,提出并验证了用考虑因素边界变化的正交设计法选取样本的可行性。研究结果表明,模型的预测精度随着正交表位级的增加而提高,即使只用3位级的正交表设计样本,也能建立预测误差低于5.7%的内燃机稳态排放特性预测模型,具有试验工作量小、简便易行的特点。 This paper studies the selection method of learning samples when the theory of neural network is used in the emission prediction model of internal combustion engine, and improves the traditional orthogonal design method according to the change of operating conditions of the internal combustion engine. The orthogonal design method with the consideration of the boundary variation is proposed and verified Feasibility of selecting samples. The results show that the prediction accuracy of the model increases with the increase of the orthogonal epitope level. Even if the sample is designed using only 3-level orthogonal table, the prediction model of the steady-state emission of the internal combustion engine can be established with the prediction error less than 5.7% , With a small test workload, easy to use features.