Abstract:Heavy diesel vehicles may experience inaccurate output values of nitrogen oxide (NOx) sensors in remote monitoring data due to ammonia slip. To effectively identify ammonia slip conditions, this study constructed a urea injection rate prediction model based on remote monitoring data items of an M3 heavy-duty diesel vehicle, where a random forest regression algorithm was used. The results indicate that there are multiple sets of highly correlated variables with Spearman correlation coefficients exceeding 0.85. Based on this, the input parameters of the model were screened. Because urea supply is aimed at controlling ammonia stor-age, and ammonia storage changes at a low frequency, the per-second urea injection rate is difficult to predict accurately. The results of predicting the average urea injection rate in different periods show that the model's prediction accuracy tends to stabi-lize when the period duration exceeds 40 seconds. Finally, the generalization ability of the prediction model was verified through the sliding window method. In all 27 windows, the coefficient of determination R2 is higher than 0.96, the mean absolute error is within 0.7%, and the root mean square error is less than 1%. The predicted urea injection rate values agree with the actual meas-ured values. |