Abstract: Heavy vehicle emission post-treatment system includes complex tail gas treatment units and supporting sensors. In order to simplify and optimize the post-treatment system, a NOx emission prediction model was established based on the improved DBSCAN algorithm and CNN model. The proposed prediction model could be deployed in the controller of the post-treatment system of heavy vehicles to simplify the sensors of the system, realize the concentration prediction function of NOx and ensure the normal operation of the post-treatment system. In view of different application scenarios, the prediction accuracy of the proposed model was analyzed through the evaluation indicators such as NOx emission concentration, Urea injection volume and NOx specific emission value. The research results show that the error between the predicted concentration of NOx emission after simplification and the measured concentration of the sensor is less than 3%. The change of Urea injection amount of the simplified post-treatment system is less than 2.08%, and the specific emission change of NOx is kept within 0.75%. And the original specific emission value of the simplified vehicle is 7.53 g/ kWh (it is 7.59g/ kWh when not simplified). The prediction of the post-treatment system embedded in the algorithm model meets the accuracy requirements.

Key words: vehicle engineering; heavy-duty vehicle; nitrogen oxide emission model; DBSCAN; CNN; after treatment system

摘要： 重型车辆的排放后处理系统，包含复杂的尾气处理单元和配套传感器。为精简后处理系统，基于改进的密度聚类算法和神经网络模型，构建氮氧化物NOx的排放预测模型，此预测模型可部署于重型车辆后处理系统的控制器中以精简系统的传感器，实现NOx的浓度预测功能，保证后处理系统的正常运转，并针对不同应用场景，通过评价指标：NOx排放浓度、Urea喷射量、NOx比排放值，来分析模型的预测精度。研究结果表明：精简后的NOx排放预测浓度与传感器测量浓度误差低于3%，精简后的后处理系统的尿素（Urea）喷射量的变化小于2.08%； NOx的比排放变化保持在0.75%以内，精简后的车辆原机比排放值为7.53 g/kWh（未精简为7.59 g/kWh）；嵌入算法模型的后处理系统的预测满足精度要求。

关键词: 车辆工程；重型车辆；氮氧化物排放模型；密度聚类；神经网络；后处理系统