在 2014 年時，WHO 提出空氣污染會對人體造成影響，其中 PM2.5 由於顆粒較小容易進入我們體內而造成傷害，對我們人體造成的損害有對我們眼睛和粘膜刺激引發過敏造成呼吸疾病，例如：引發氣喘，也會對我們神經系統有所損傷。此外根據研究 [26] 長期暴露在 PM2.5 濃度較高的環境中，帶來的死亡風險也會上升，因此越來越受到人們關注。PM2.5為一時空間資料，且生活中影響 PM2.5 濃度的因素眾多，各因素之間存在複雜的影響關係，因此預測 PM2.5 濃度值的研究會面臨的挑戰是要如何取得眾多影響因素之間的時空間關係。
本論文主要目標是針對 PM2.5 做預測，而目前世界各地針對空氣品質問題架設了許多站點來監測數值，但站點的密度不均勻。本論文的的研究目的有兩個，第一個是預測現有站點未來 48 小時每個小時 PM2.5 濃度值，第二個是預測無站點處未來 48 小時每個小時 PM2.5 濃度值。透過 PM2.5 預測，人們可以根據預測結果來做決策，來減少人們長期暴露在 PM2.5 濃度過高的環境中，減少 PM2.5 對人體的傷害。近幾年，圖神經網路的盛行，過去有許多研究利用圖神經網路來預測時空間資料的問題，本研究設計一結合時間專注力機制之遞歸圖神經網路，並利用Encoder以及Decoder機制來處理Sequence資料，使神經網路能更有效的預測 PM2.5 數值。本研究提出之圖神經網路架構不但可以預測現有站點未來PM2.5 濃度值，亦可以預測任意非站點處未來 PM2.5 濃度值。我們訓練圖神經網路所使用的空氣品質歷史資料為台灣行政院環保署架設在台灣各地的站點資料，以及使用了中央氣象局架設的自動氣象站收集的天氣資料，前述兩類型資料為時空間資料。此外針對空間關係，我們利用 Open Street Map 獲取各地點周邊各類型 POIs 數量，使得圖神經網路依據不同類型的 POIs 分佈來學習各站點的空間關係。
在實驗中本研究使用台灣真實資料來驗證本研究提出之架構有效性，並比較了不同特徵資料對於預測的影響以及分析架構設計中各元件的貢獻。此外，針對非站點處的預測，本研究也使用了真實數據來做驗證，並分析架構設計對於預測結果的影響。

In 2014, WHO proposed that air pollution would have an impact on the human body. Among these air pollutions, PM2.5 particles are small and easily enter our bodies and cause harm. The damage to our human body includes irritation to our eyes and mucous membranes, causing allergies and respiratory diseases, such as causing asthma. It will also damage our nervous system. In addition, according to research [26], long-term exposure to an environment with a high PM2.5 concentration will increase the risk of death. That is why people are paying more and more attention to the issue of air pollution. PM2.5 is temporal-spatial data. Many factors are influencing PM2.5 concentration in life, and there are complex influence relationships among various factors. Therefore, the challenge for the study of predicting the concentration of PM2.5 is how to obtain the temporal and spatial relationships among many influencing factors.
The main goal of this paper is to predict PM2.5. At present, many stations have been set up to monitor the air quality around the world, but the density of the stations is uneven. There are two research purposes in this paper. The first is to predict the PM2.5 concentration value of each hour in the next 48 hours of the existing station, and the other one is to predict the PM2.5 concentration value of each hour in the next 48 hours in any place without stations. Through PM2.5 prediction, people can make decisions based on the predicted results to reduce long-term exposure to an environment with excessive PM2.5 concentration, thus reducing the harm of PM2.5 to the human body. In recent years, graph neural networks have become popular. In the past, there have been many studies on the use of graph neural networks to predict temporal and spatial data. We design a recurrent graph neural network that combines temporal attention mechanisms and use Encoder and Decoder mechanisms to process the sequence data so that the model can predict the PM2.5 value more effectively. The graph neural network architecture proposed in this study can not only predict the future PM2.5 concentration value of the existing station but also predict the future PM2.5 concentration value of any non-station place. We use data from air quality stations set up by the Taiwan Environmental Protection Administration Executive Yuan to train the graph neural network, and we also used the weather data collected by the automatic weather station set up by the Taiwan Central Weather Bureau. These two types of data are both temporal-spatial data. In addition, for the spatial relationship, we use Open Street Map to obtain the number of various types of POIs around each location, so that the graph neural network could learn the spatial relationship of each station based on the distribution of different types of POIs.
In the experiment, we use real data from Taiwan to verify the effectiveness of the architecture proposed in this study. We also compared the impact of different feature data on the prediction and analyzed the contribution of each component in the architectural design. In addition, for non-station place predictions, we also use real data for verification and analysis of the impact of architecture design on the prediction results.

[1] Bahdanau, D., Cho, K.H. and Bengio, Y. 2015. Neural machine translation by jointly learning to align and translate. 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings (2015).
[2] Cheng, W., Shen, Y., Zhu, Y., and Huang, L. A Neural Attention Model for Urban Air Quality Inference: Learning the Weights of Monitoring Stations.
[3] Cobourn, W.G. 2007. Accuracy and reliability of an automated air quality forecast system for ozone in seven Kentucky metropolitan areas. Atmospheric Environment. (2007). DOI:https://doi.org/10.1016/j.atmosenv.2007.03.024.
[4] Díaz-Robles, L.A., Ortega, J.C., Fu, J.S., Reed, G.D., Chow, J.C., Watson, J.G. and Moncada-Herrera, J.A. 2008. A hybrid ARIMA and artificial neural networks model to forecast particulate matter in urban areas: The case of Temuco, Chile. Atmospheric Environment. (2008). DOI:https://doi.org/10.1016/j.atmosenv.2008.07.020.
[5] Du, S., Li, T., Yang, Y. and Horng, S.J. 2018. Deep Air Quality Forecasting Using Hybrid Deep Learning Framework. arXiv.
[6] Gers, F.A., Schmidhuber, J. and Cummins, F. 1999. Learning to forget: Continual prediction with LSTM. IEEE Conference Publication (1999).
[7] Goyal, P., Chan, A.T. and Jaiswal, N. 2006. Statistical models for the prediction of respirable suspended particulate matter in urban cities. Atmospheric Environment. (2006). DOI:https://doi.org/10.1016/j.atmosenv.2005.11.041.
[8] Hochreiter, S. and Schmidhuber, J. 1997. LSTM 1997. Neural Computation. 9, 8, November 15, 1997 (1997).
[9] Hoi, K.I., Yuen, K. V. and Mok, K.M. 2008. Kalman filter based prediction system for wintertime PM10 concentrations in Macau. Global Nest Journal. (2008). DOI:https://doi.org/10.30955/gnj.000545.
[10] Hsieh, H.P., Lin, S. De and Zheng, Y. 2015. Inferring air quality for station location recommendation based on urban big data. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2015).
[11] https://airtw.epa.gov.tw/ENG/Information/Standard/AirQualityIndicator.aspx: .
[12] Kumar, U. and Jain, V.K. 2010. ARIMA forecasting of ambient air pollutants (O 3, NO, NO 2 and CO). Stochastic Environmental Research and Risk Assessment. (2010). DOI:https://doi.org/10.1007/s00477-009-0361-8.
[13] Le, V.D., Bui, T.C. and Cha, S.K. 2020. Spatiotemporal deep learning model for citywide air pollution interpolation and prediction. Proceedings - 2020 IEEE International Conference on Big Data and Smart Computing, BigComp 2020 (2020).
[14] Li, L., Zhang, X., Holt, J.B., Tian, J. and Piltner, R. 2011. Spatiotemporal interpolation methods for air pollution exposure. SARA 2011 - Proceedings of the 9th Symposium on Abstraction, Reformulation, and Approximation (2011).
[15] Li, X., Peng, L., Hu, Y., Shao, J. and Chi, T. 2016. Deep learning architecture for air quality predictions. Environmental Science and Pollution Research. (2016). DOI:https://doi.org/10.1007/s11356-016-7812-9.
[16] Liang, Y., Ke, S., Zhang, J., Yi, X. and Zheng, Y. 2018. Geoman: Multi-level attention networks for geo-sensory time series prediction. IJCAI International Joint Conference on Artificial Intelligence (2018).
[17] Lim, S.Y., Chin, L.Y., Mah, P. and Wee, J. 2008. Arima and Integrated Arfima Models for Forecasting Air Pollution Index in Shah Alam , Selangor. The Malaysian Journal of Analytical Science. (2008).
[18] Lin, Y., Mago, N., Gao, Y., Li, Y., Chiang, Y.Y., Shahabi, C. and Ambite, J.L. 2018. Exploiting spatiotemporal patterns for accurate air quality forecasting using deep learning. GIS: Proceedings of the ACM International Symposium on Advances in Geographic Information Systems (2018).
[19] Luo, Z., Huang, J., Hu, K., Li, X. and Zhang, P. 2019. Accuair: Winning solution to air quality prediction for KDD Cup 2018. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2019).
[20] Ma, J., Ding, Y., Gan, V.J.L., Lin, C. and Wan, Z. 2019. Spatiotemporal Prediction of PM2.5 Concentrations at Different Time Granularities Using IDW-BLSTM. IEEE Access. (2019). DOI:https://doi.org/10.1109/ACCESS.2019.2932445.
[21] McKendry, I.G. 2002. Evaluation of artificial neural networks for fine particulate pollution (PM10 and PM2.5) forecasting. Journal of the Air and Waste Management Association. (2002). DOI:https://doi.org/10.1080/10473289.2002.10470836.
[22] Perez, P. and Salini, G. 2008. PM2.5 forecasting in a large city: Comparison of three methods. Atmospheric Environment. (2008). DOI:https://doi.org/10.1016/j.atmosenv.2008.07.035.
[23] Pérez, P., Trier, A. and Reyes, J. 2000. Prediction of PM2.5 concentrations several hours in advance using neural networks in Santiago, Chile. Atmospheric Environment. (2000). DOI:https://doi.org/10.1016/S1352-2310(99)00316-7.
[24] Qi, Y., Li, Q., Karimian, H. and Liu, D. 2019. A hybrid model for spatiotemporal forecasting of PM 2.5 based on graph convolutional neural network and long short-term memory. Science of the Total Environment. (2019). DOI:https://doi.org/10.1016/j.scitotenv.2019.01.333.
[25] Qi, Z., Wang, T., Song, G., Hu, W., Li, X. and Zhang, Z. 2018. Deep Air Learning: Interpolation, Prediction, and Feature Analysis of Fine-Grained Air Quality. IEEE Transactions on Knowledge and Data Engineering. 30, 12 (2018). DOI:https://doi.org/10.1109/TKDE.2018.2823740.
[26] Schwartz, J. and Neas, L.M. 2000. Fine particles are more strongly associated than coarse particles with acute respiratory health effects in schoolchildren. Epidemiology. 11, 1 (2000). DOI:https://doi.org/10.1097/00001648-200001000-00004.
[27] Sharma, E., Deo, R.C., Prasad, R., Parisi, A. V. and Raj, N. 2020. Deep Air Quality Forecasts: Suspended Particulate Matter Modelling with Convolutional Neural and Long Short-Term Memory Networks. IEEE Access. (2020). DOI:https://doi.org/10.1109/access.2020.3039002.
[28] Singh, K.P., Gupta, S. and Rai, P. 2013. Identifying pollution sources and predicting urban air quality using ensemble learning methods. Atmospheric Environment. (2013). DOI:https://doi.org/10.1016/j.atmosenv.2013.08.023.
[29] Soh, P.W., Chang, J.W. and Huang, J.W. 2018. Adaptive Deep Learning-Based Air Quality Prediction Model Using the Most Relevant Spatial-Temporal Relations. IEEE Access. 6, (2018). DOI:https://doi.org/10.1109/ACCESS.2018.2849820.
[30] Suárez Sánchez, A., García Nieto, P.J., Iglesias-Rodríguez, F.J. and Vilán Vilán, J.A. 2013. Nonlinear air quality modeling using support vector machines in Gijón urban area (Northern Spain) at local scale. International Journal of Nonlinear Sciences and Numerical Simulation. (2013). DOI:https://doi.org/10.1515/ijnsns-2012-0119.
[31] Suárez Sánchez, A., García Nieto, P.J., Iglesias-Rodríguez, F.J. and Vilán Vilán, J.A. 2013. Nonlinear air quality modeling using support vector machines in Gijón urban area (Northern Spain) at local scale. International Journal of Nonlinear Sciences and Numerical Simulation. 14, 5 (Aug. 2013), 291–305. DOI:https://doi.org/10.1515/ijnsns-2012-0119.
[32] Sutskever, I., Vinyals, O. and Le, Q. V. 2014. Sequence to sequence learning with neural networks. Advances in Neural Information Processing Systems (2014), 3104–3112.
[33] Vardoulakis, S., Fisher, B.E.A., Pericleous, K. and Gonzalez-Flesca, N. 2003. Modelling air quality in street canyons: A review. Atmospheric Environment. (2003). DOI:https://doi.org/10.1016/S1352-2310(02)00857-9.
[34] Wang, S., Meng, Q., Li, Y., Meng, L., Zhang, J. and Gao, F. 2020. PM2.5-GNN: A Domain Knowledge Enhanced Graph Neural Network For PM2.5 Forecasting. arXiv.
[35] Wu, Z., Wang, Y. and Zhang, L. 2019. MSSTN: Multi-Scale Spatial Temporal Network for Air Pollution Prediction. Proceedings - 2019 IEEE International Conference on Big Data, Big Data 2019 (Dec. 2019), 1547–1556.
[36] Xu, K., Hu, W., Leskovec, J. and Jegelka, S. 2018. How powerful are graph neural networks? arXiv.
[37] Xu, Z., Kang, Y., Cao, Y., Neural, Z.L.-I.T. on and 2020, undefined Spatiotemporal Graph Convolution Multifusion Network for Urban Vehicle Emission Prediction. ieeexplore.ieee.org.
[38] Yi, X., Zhang, J., Wang, Z., Li, T. and Zheng, Y. 2018. Deep distributed fusion network for air quality prediction. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2018).
[39] Yu, B., Yin, H. and Zhu, Z. 2018. Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting. IJCAI International Joint Conference on Artificial Intelligence (2018), 3634–3640.
[40] Zhang, Y., Wang, Y., Gao, M., Ma, Q., Zhao, J., Zhang, R., Wang, Q. and Huang, L. 2019. A Predictive Data Feature Exploration-Based Air Quality Prediction Approach. IEEE Access. (2019). DOI:https://doi.org/10.1109/ACCESS.2019.2897754.
[41] Zhang, Z., Cui, P. and Zhu, W. 2018. Deep Learning on Graphs: A Survey. arXiv.
[42] Zheng, Y., Liu, F. and Hsieh, H.P. 2013. U-Air: When urban air quality inference meets big data. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2013).
[43] Zheng, Y., Yi, X., Li, M., Li, R., Shan, Z., Chang, E. and Li, T. 2015. Forecasting fine-grained air quality based on big data. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2015).
[44] Zhou, J., Cui, G., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C. and Sun, M. 2018. Graph Neural Networks: A Review of Methods and Applications. arXiv.
[45] Zhou, X., Huang, W., Zhang, N., Hu, W., Du, S., Song, G. and Xie, K. 2015. Probabilistic dynamic causal model for temporal data. Proceedings of the International Joint Conference on Neural Networks (Sep. 2015).