Deep Spatial-Temporal Fusion network for Fine-grained Air Quality Prediction

The prediction of spatially fine-grained air quality is an important direction in urban air computing. Solving the problem can provide useful information for urban environmental governance and residents' health improvement. This paper proposes a general approach to solve the problem, which consists of data completion component, similar region selection component, and a deep spatial-temporal fusion network(DSTFN). Considering the missing of historical air quality data, the tensor decomposition method is used in the data completion component. Considering the similarity of air quality between urban regions, the similar region selection component uses heterogeneous data to calculate the spatial similarity between regions. The deep spatial-temporal fusion network fuse urban heterogeneous data to predict air quality for simultaneously capturing the affecting factors. We evaluated our approach on real data sources obtained in Beijing, and the experimental results show its advantages over baseline methods.