关键词: 空气质量, 深度学习, 时空特征, 多粒度, 突变

Abstract: Abstract: Air quality data, as a typical spatio-temporal data, exhibits complex multi-scale intrinsic characteristics and has sudden change problem. Concern the problem that existing air quality prediction methods perform poorly when dealing with air quality prediction tasks containing a large amount of sudden changes, a multi-granularity mutation fitting network (MACFN) for air quality prediction was proposed. Specifically, multiscale feature extraction was first performed on the input data since the air quality data has temporal periodicity. Then, a graph convolution network and a temporal convolution network are used to extract the spatial correlation and temporal dependence of the air quality data, respectively. Finally, to reduce the prediction error, an abrupt change fitting network was designed to adaptively learn the abrupt change part of the data. The proposed model was experimentally evaluated on three real air quality datasets, and the RMSE decreases by about 11.6%, 6.3%, and 2.2% on the three datasets, respectively, when compared to the suboptimal model MSSTN. The experimental results show that the MACFN model was able to efficiently capture complex spatio-temporal relationships and perform better in the task of predicting air quality that was prone to sudden changes with a large magnitude of variability.

Key words: air quality prediction, deep learning, spatio-temporal, multi-grain size, abrupt change