Predicting the transient NOx emissions of the diesel vehicle based on LSTM neural networks

Abstract

Nitrogen oxide (NOx) emissions play an important role in the study of diesel engine pollutant emissions. This study introduces the long short-term memory (LSTM) neural network to estimate the transient NOx emissions of diesel vehicles. The LSTM deep neural network is used to build the prediction model to ensure the stability as well as the accuracy of the model. The results show that the model has better predictive performance and stability than the two commonly used benchmark models, and the following conclusions are drawn: (1) LSTM has better learning and prediction ability for transient changes in NOx emissions. Compared to prediction with random forest (RF) and support vector regression (SVR), the mean absolute deviation and root mean square error of LSTM are reduced by about 23.6% and 8.3% at least, which also indicated that the input parameters selection method was effective. (2) LSTM is a general estimation approach for time series data, which can reduce the suppression effect of transient data changes on model prediction, and has high prediction accuracy, and can be employed for real road NOx emission analysis.