A spatial correlation prediction model of urban <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" altimg="si5.svg"><mml:mrow><mml:msub><mml:mrow><mml:mi mathvariant="normal">PM</mml:mi></mml:mrow><mml:mrow><mml:mn>2.5</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:math> concentration based on deconvolution and LSTM
Bo Zhang, Yuan Liu, Ruihan Yong, Guojian Zou, Ru Yang, Jianguo Pan, Maozhen Li
Abstract
Precise prediction of air pollutants can effectively reducre the occurrence of heavy pollution incidents. With the current surge of massive data, deep learning appears to be a promising technique to achieve dynamic prediction of air pollutant concentration from both the spatial and temporal dimensions. This paper presents Dev-LSTM, a prediction model building on deconvolution and LSTM. The novelty of Dev-LSTM lies in its capability to fully extract the spatial feature correlation of air pollutant concentration data, preventing the excessive loss of information caused by traditional convolution. At the same time, the feature associations in the time dimension are mined to produce accurate prediction results. Experimental results show that Dev-LSTM outperforms traditional prediction models on a variety of indicators.