Abstract: Selective catalytic reduction (SCR) technology with urea as reducing agent is the main way to reduce NOx emissions of diesel vehicles, and accurate prediction of NOx is the premise to achieve accurate control of SCR. In this paper, a method of random forest combined with gate recurrent unit (GRU) is proposed to predict NOx. In view of the complexity of NOx generation factors, random forest is used to select features that have a large impact on the results, and the NOx prediction model based on gate unit is constructed. The experimental results show that the mean square error of the prediction results using random forest combined with GRU under transient and steady-state conditions is 66.419×10-6 and 63.423×10-6 , which proves that the model has high accuracy and good generalization ability.

Key words: NOx prediction; Neural network; Diesel engine; Random forest; GRU algorithm

摘要： 以尿素为还原剂的选择性催化还原（SCR）技术是降低柴油车 NOx 排放的主要途径， 而实现 NOx 的精准预测是实现对 SCR 精准控制的前提，文章提出使用随机森林结合门控循 环单元（GRU）的方法对 NOx 进行预测。针对 NOx 生成因素的复杂性，使用随机森林进行 特征选取，选取对结果影响大的特征，并构建基于优化门控单元的 NOx 的预测模型。实验结 果显示，在瞬态与稳态工况下使用随机森林结合 GRU 的预测结果的均方误差分别为 66.419×10-6 与 63.423×10-6，证明模型具有较高的精准度以及良好的泛化能力。

关键词: NOx 预测；神经网络；柴油机；随机森林；GRU 算法