基于组合模型的高原环境GDI汽油车排放预测

Emission Prediction of GDI Vehicle Based on Combination Model under Plateau Environment

为预测高原环境下缸内直喷(Gasoline Direct Injection, GDI)汽油车CO和PN的瞬时排放量,开发并评估了一套基于深度学习的排放预测模型。利用便携式车载排放测试系统对一辆GDI汽油车进行实际道路排放测试;加入奇异谱分析对原始时间序列进行处理,剔除时间序列中的异常值;利用XGBoost模型对GDI汽油车的CO和PN的瞬时排放进行初步预测,并利用SVR模型进行残差修正得到最终的预测结果。将预测结果与实际道路排放试验中使用PEMS设备测量的实际值进行比较,结果表明,XGBoost-SVR排放预测模型能较好地预测GDI汽油车瞬时CO和PN的排放,相比单一的XGBoost模型,RMSE分别提高了22.9%和39.7%,决定系数R2均大于0.9,支持预测结果的可靠性。该模型对监测高原环境下GDI汽油车实际道路排放具有一定的工程意义。

In order to predict the transient CO and PN emissions of gasoline direct injection(GDI)vehicle in plateau,a prediction model based on deep-learning was developed and assessed.First,a portable emission measurement system(PEMS)was used to test the road emission of a GDI vehicle.The singular spectrum analysis is introduced to process the original time series and eliminate the outliers of the time series.The XGBoost model was used to preliminary predict CO and PN emissions of GDI vehicle,the SVR model was then used to correct the residual,and the final predicted values were obtained.Finally,the predicted results were compared with experimental values of road emission measured using PEMS.The experimental results show that the established XGBoost-SVR emission prediction model can better predict the transient CO and PN emissions of GDI vehicle.Compared with the single XGBoost model,the RMSE improves by 22.9%and 39.7% respectively.R 2 determination coefficients are both larger than 0.9,supporting the reliability of predicted results.This model has certain engineering significance for monitoring the emissions of GDI vehicles under situations of actual road driving in domestic plateau environment.