基于重型柴油车远程监控数据的尿素喷射速率预测

Prediction of Urea Injection Rates Based on Remote Monitoring Data of Heavy-Duty Diesel Vehicles

基于一辆M3类重型柴油车的远程监控数据项,构建了尿素喷射速率预测模型,采用随机森林回归算法进行预测,对远程监控数据中缺少的尿素喷射速率的预测进行了研究。结果表明远程监控数据中存在多组高度相关的变量,其Spearman相关系数均超过0.85,基于此对模型的输入参数进行筛选。由于尿素供给以控制氨存储为目标,而氨存储的变化频率较低,逐秒的尿素喷射速率难以精确预测。不同周期内平均尿素喷射速率的预测结果显示,模型预测精度趋于稳定。当周期超过40 s后,预测精度趋于稳定。通过滑动窗口法验证了预测模型的泛化能力,在全部27个窗口中,决定系数R~2普遍高于0.96,平均绝对误差均控制在0.7%以内,均方根误差基本上低于1%,尿素喷射速率预测值与实际测量值吻合良好。

Based on the remote monitoring data of a M3 heavy-duty diesel vehicle,a urea injection rate prediction model was constructed,and the prediction of urea injection rate missing in the remote monitoring data was studied by using the random forest regression algorithm.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 storage,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 stabilize 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 measured values.