Traffic emission estimation under incomplete information with spatiotemporal convolutional GAN

Precise emission estimation of vehicle exhausts is crucial to urban traffic pollution prevention and control. Existing methods utilize the widely distributed and large number of GPS data to estimate the emission distribution of vehicles in the road network. However, the emission features are insufficient to decrease the estimation accuracy when the data distribution is uneven and sparse in the spatial and temporal domains. To address this problem, we propose a two-step emission estimation model under incomplete information, which exploits the spatiotemporal propagation features of emission information. Specifically, an adaptive smoothing strategy reconstructs a second-by-second emission rate field by modeling the correlation of neighboring traffic states, to address the inconsistency of time intervals between emission models and sampling intervals. Then a spatiotemporal convolutional GAN (ST-CGAN) is proposed, which introduces the spatiotemporal convolution operator to generate the traffic emission by temporal features and structural similarity. We evaluated the proposed method using the GPS trajectory data on Didi Chuxing GAIA Open Dataset. The framework aligns GPS data and emission models and reconstructs effectively the high-emission features in traffic networks. The proposed ST-CGAN generates a more reasonable spatial and temporal distribution of vehicle emissions than state-of-the-art methods.