A spatial correlation prediction model of urban PM2.5 concentration based on deconvolution and LSTM

• DOI: 10.1016/j.neucom.2023.126280
• Publication Date: 2023-04-28T08:46:10Z
• Authors: Bo Zhang; Yuan Liu; RuiHan Yong; Guojian Zou; Ru Yang; Jianguo Pan; Maozhen Li
• Keywords: Predictive modelling; Feature (linguistics); Convolution (computer science)

Precise prediction of air pollutants can effectively reducre the occurrence heavy pollution incidents. With current surge massive data, deep learning appears to be a promising technique achieve dynamic pollutant concentration from both spatial and temporal dimensions. This paper presents Dev-LSTM, model building on deconvolution LSTM. The novelty Dev-LSTM lies in its capability fully extract feature correlation preventing excessive loss information caused by traditional convolution. At same time, associations time dimension are mined produce accurate results. Experimental results show that outperforms models variety indicators.