A graph-based LSTM model for PM2.5 forecasting

Abstract Accuracy prediction of air quality is of crucial importance for people to take precautions and improve environmental conditions. By introducing adjacency matrix in Long Short-Term Memory (LSTM) cell, we propose in this research a Graph-based Long Short-Term Memory (GLSTM) model to predict PM2.5 concentration in Gansu Province of Northwest China. We regard all air quality monitoring stations as a graph, and construct a parameterized adjacency matrix on the basis of the adjacency matrix of the graph. Through the combination of parameterized adjacency matrix and LSTM, we introduce spatiotemporal information to achieve PM2.5 prediction. The advantage of GLSTM is that it can realize synchronous operation of all stations, making it unnecessary to train different model for each monitoring station to obtain the overall PM2.5 variation of a certain area. The parameterized adjacency matrix also enhances the interpretability of the model. By visualizing the parameterized adjacency matrix obtained from the end-to-end PM2.5 prediction task in training, the importance of introducing spatial information, i.e. the distribution importance of surrounding stations to a specific station is clearly demonstrated. We compared our model with several newly reported methods, and found that it achieved the best results on PM2.5 prediction tasks at almost all stations, which proved the effectiveness of the GLSTM model.

[1]  H. Akimoto,et al.  Future prediction of surface ozone over east Asia using Models-3 Community Multiscale Air Quality Modeling System and Regional Emission Inventory in Asia , 2008 .

[2]  Cyrus Shahabi,et al.  Exploiting spatiotemporal patterns for accurate air quality forecasting using deep learning , 2018, SIGSPATIAL/GIS.

[3]  D. Byun,et al.  Review of the Governing Equations, Computational Algorithms, and Other Components of the Models-3 Community Multiscale Air Quality (CMAQ) Modeling System , 2006 .

[4]  Thomas O Talbot,et al.  Assessing the impact of fine particulate matter (PM2.5) on respiratory-cardiovascular chronic diseases in the New York City Metropolitan area using Hierarchical Bayesian Model estimates. , 2016, Environmental research.

[5]  Feng Liu,et al.  The forecasting research of early warning systems for atmospheric pollutants: A case in Yangtze River Delta region , 2015 .

[6]  Jun Ma,et al.  Deep learning-based PM2.5 prediction considering the spatiotemporal correlations: A case study of Beijing, China. , 2020, The Science of the total environment.

[7]  Binxu Zhai,et al.  Development of a stacked ensemble model for forecasting and analyzing daily average PM2.5 concentrations in Beijing, China. , 2018, The Science of the total environment.

[8]  Jiangshe Zhang,et al.  A novel method for identifying hotspots and forecasting air quality through an adaptive utilization of spatio-temporal information of multiple factors. , 2020, The Science of the total environment.

[9]  Vishal M. Patel,et al.  HA-CCN: Hierarchical Attention-Based Crowd Counting Network , 2019, IEEE Transactions on Image Processing.

[10]  Z. Meng,et al.  Impact of PM2.5 derived from dust events on daily outpatient numbers for respiratory and cardiovascular diseases in Wuwei, China , 2013 .

[11]  Xiyuan Chen,et al.  Spatiotemporal prediction of air quality based on LSTM neural network , 2021 .

[12]  Nitesh V. Chawla,et al.  Heterogeneous Graph Neural Network , 2019, KDD.

[13]  Yanran Li,et al.  PM2.5-GNN: A Domain Knowledge Enhanced Graph Neural Network For PM2.5 Forecasting , 2020, SIGSPATIAL/GIS.

[14]  Max Welling,et al.  Semi-Supervised Classification with Graph Convolutional Networks , 2016, ICLR.

[15]  Robert Vautard,et al.  A comparison of simulated and observed ozone mixing ratios for the summer of 1998 in Western Europe , 2001 .

[16]  Yo-Ping Huang,et al.  An LSTM-based aggregated model for air pollution forecasting , 2020 .

[17]  Zhanxing Zhu,et al.  Spatio-temporal Graph Convolutional Neural Network: A Deep Learning Framework for Traffic Forecasting , 2017, IJCAI.

[18]  Sumit Sharma,et al.  Air quality forecasting using artificial neural networks with real time dynamic error correction in highly polluted regions. , 2020, The Science of the total environment.

[19]  Li-Chiu Chang,et al.  Multi-output support vector machine for regional multi-step-ahead PM2.5 forecasting. , 2019, The Science of the total environment.

[20]  Il Kyu Eom,et al.  Fast Single Image Dehazing Using Saturation Based Transmission Map Estimation , 2020, IEEE Transactions on Image Processing.

[21]  Gordon Reikard.  Forecasting volcanic air pollution in Hawaii: Tests of time series models , 2012 .

[22]  Chun-Xia Zhang,et al.  Improving text classification with weighted word embeddings via a multi-channel TextCNN model , 2019, Neurocomputing.

[23]  Qi Li,et al.  A hybrid model for spatiotemporal forecasting of PM2.5 based on graph convolutional neural network and long short-term memory. , 2019, The Science of the total environment.

[24]  Gregory R. Herman,et al.  Forecasting Severe Weather with Random Forests , 2020, Monthly Weather Review.