Mobile source emission model based on temporal features transfer*

Abstract

To address the problems of low prediction accuracy and poor stability caused by dynamic changes in data distribution in the process of mobile source pollution time series prediction, this paper establishes a mobile source emission prediction model (TFT_GRU) with temporal transfer for mobile source emission prediction. First, the continuous time series data are divided into multiple sub-segments of the distribution with maximum variability, and the time series data are divided into multiple sub-segments to be subjected to temporal feature transfer. Then the TFT_GRU model with temporal invariance is obtained by adding the disparity measure to the loss function of the base RNN model and performing iterative optimization. Finally, experiments are conducted on the OBD dataset of diesel vehicle monitoring in Hefei City on June 9, 2020, and the feasibility and effectiveness of the proposed model in mobile source pollution prediction are verified by comparing with other temporal models.