Emission modeling for new-energy buses in real-world driving with a deep learning-based approach

Abstract Nowadays, new energy bus is gradually replacing those with diesel engines with its better environmental protection characteristics. As one of the main types of new energy buses, liquefied natural gas (LNG) bus still emits a large number of pollutants as well as greenhouse gases. Consequently, the primary objective of this study is to investigate how the deep learning approach contributes to emission prediction by leveraging emission datasets of LNG buses. This study uses datasets of LNG buses measured in Zhenjiang, China to illustrate the procedure. A deep learning framework based on the gated recurrent unit (GRU) capturing both time dependence and external factors including speed, acceleration, dynamic passenger load, and road grade is proposed for real-time emission rates modeling. The results suggest that the predicted values are very close to the true values except that the estimation errors are relatively large when the measured emission rates are near 0. Moreover, one commonly used statistical model and two state-of-art machine learning models are selected as benchmark methods to compare with the proposed model for the emission modeling tasks. The comparative analyses suggest that the proposed GRU-based emission model outperforms the benchmark approaches for different kinds of emission prediction tasks in terms of higher prediction accuracy and training efficiency. The results indicate that the proposed model has a good ability to predict exhaust emissions for LNG buses in real-world driving, which can potentially guide transportation and environmental protection engineers to select appropriate methods for monitoring urban traffic emissions.