基于机器学习的柴油机碳烟颗粒质量排放预测模型

基于SC7H涡轮增压柴油机试验台架,开展了非道路瞬态试验循环下的柴油机排放试验,研究了瞬态循环的工况对碳烟颗粒质量浓度的影响。收集与碳烟颗粒质量浓度相关的各类传感器数据,构建一个大型的柴油机碳烟排放数据集。构建LGB梯度树模型和循环神经网络模型,采用数据集对它们进行训练,然后采用自学习算法对两种模型进行融合,获得一个更高准确度的预测碳烟质量排放融合模型。预测与实测结果的比较表明,构建的融合模型能较为准确地预测柴油机排放的即DPF入口的碳烟质量浓度实时变化,为柴油机后处理过程中碳载量的准确计算以及控制策略的开发提供参考。

A Prediction Model for Diesel Engine Soot Mass Emissions Based on Machine Learning

The emission test over the non-road transient cycle was carried out in a test bench for the SC7H turbocharged diesel engine. The influence of transient cycle conditions on the mass concentration of soot particles was studied. The various sensor data related to the mass concentration of soot particles were collected to construct the training dataset. After that, a LGB gradient tree model and a recurrent neural network model were constructed respectively. Additionally the dataset was used to train the models, and a self-learning algorithm was applied to fuse the two models into a more accurate fusion model for predicting soot mass emission. The comparison with the experimental results shows that the fusion model more accurately predicts the real-time change of the Diesel Particulate Filter (DPF) soot mass concentration,which provides a reference for soot-loading estimation and control strategy development.