近年來，空氣中的細懸浮微粒漸漸地引起人們的關注，其直徑等於或小於2.5μg/m^3，甚至只有頭髮的直徑的1/30倍，因為它太小難以阻擋，可以穿過肺泡微血管透或血液循環到人體全身，對人體的健康有重大的影響，因此對人們來說監測與預測空氣品質成為非常重要的事情。現今我們可以透過安裝一個小型的空氣檢測儀或是上網查詢就能輕易地得到當下的空氣品質資訊，但我們更需要知道未來的空氣品質以方便規劃未來的行程。此研究致力於預測未來的空氣品質，過去的文獻指出城市中的空氣品質是非線性的，也因為很多因素而變化，像是溫度、濕度、風速、風向、雨量、地理位置，至此我們運用空氣盒子提供的空氣品質資料與氣象局提供的天氣資料作結合，以時間性與空間性的資料來預測未來的空氣品質。空氣品質是連續性的資料，在同個地點若是上的時間點監測到的空氣品質良好，那下個時間點的空氣品質也會有非常高的可能性也良好；而同個地點的空氣也可能會有時間性上的相似，例如季節性的影響；而鄰近地點的空氣品質也有高可能性會監測到類似的結果，例如測站A周圍的測站都監測到良好的空氣品質，那測站A也很可能會監測到良好的空氣品質。因此我們找尋與欲預測資料時間上最近的監測資料，與欲預測資料氣象因素最相似的過去資料，透過類似加權平均的評估方法，以天氣維度的差異程度當作權，預測出初步的空氣品質，而後在將初步的預測值和周圍測站同時間的預測值再做一次加權平均，最後在透過鄰近測站的之間以類似於正規化的方式使預測的空氣品質成果趨勢相似，得到最終預測出的空氣品質。為了使空氣品質預測可以有更好的結果，本研究也設計了一個新測站建置地點的推薦，此方法是藉由迭代方法最小化預測結果的不確定性，我們利用只監測到的資料計算出新測站建置候選地點的不確定性，不確定性代表候選點周圍的測站氣象因素相似，但是預測出的空氣品質不相似，也因此我們認為不確定性大的候選地點需要建置新測站，提供周圍測站參考，降低預測的不確定性，我們以候選地點附近的空氣品質測站的預測值來估算每個候選地點的不確定性，選出不確定性最大的候選地點作為第一推薦的建置地點。然後被選取的地點會被當作有測站的已知資料加入空氣品質的預測，計算出剩下候選地點的不確定性，一樣在挑出不確定性最大的地點當作第二推薦的地點，我們會重複這樣的步驟直到所有候選地點被排序完成，得到一個完整的新測站建置地點推薦順序。與知名方法皮爾森相關係數與空間內插的反距離權重法相比，我們的方法表現較佳，且在預測較久的未來時較為穩定。但是在新測站建置推薦上，我們的方法卻不如欲如預期。

Recently, air pollution has been mainly induced by particle matters becomes an important issue, where the diameter is less than or equal to 2.5 μg/m^3. It is so tiny that it can go through alveolar microvascular and enter our body. Therefore, PM2.5 will make a significant impact on human health. Therefore, monitoring and forecasting the air quality is an indispensable task for human society. Nowadays, we can easily acquire Air Quality Indices (AQIs) by installing a small-scale air quality sensor or downloading from some freely authorized databases. However, people demand farther PM2.5 information to plan their route. This research aims to forecast PM2.5 value in the future hours. Previous studies indicated that the air quality varies nonlinearly in urban areas and depends on several factors such as temperature, humidity and wind speed. Therefore, we combined air quality data from AirBox and meteorology data from Center Weather Bureau and forecast PM2.5 value in temporal and spatial aspects. Air quality is a continuous data. If monitored air quality is good at the last time stamp, the next monitored air quality has high possibility to be good at the same location. And air quality may have some regular in the history data. Besides, a device has high possibility to monitor good air quality when devices around monitor good air quality at the same time. Therefore, we choose some recent, similar and near data to forecast PM2.5 for each device we want to forecast. We forecast PM2.5 values via the algorithm similar to weighted average method. It can figure out the time intervals with similar weather condition. Then, forecast PM2.5 values from those stations adjacent to the target place are used to make the result trends of previous step similar. Finally, the error is calculated to examine the accuracy of our method. We also propose a new device allocation recommendation to improve the result of forecast accuracy. We design an iterative method to decrease uncertainty of forecast. Uncertainty is the degree of PM2.5 values difference of near devices which have similar meteorology feature. We consider that candidate location with high uncertainty demand a device to be referred by near devices. Through evaluation of the difference of near devices, we suggest the best location to install a new device. The chosen location will be regarded as a location with a device to forecast PM2.5 and evaluate uncertainty of the rest candidate locations. The candidate with the highest uncertainty will be chosen as the second location we recommend. These steps are repeated until all of candidate locations are listed. Finally, a complete recommendation list will be produced. In contrast to some famous method, including Pearson’s Correlation Coefficient and Inverse Distance Weighted, our method preforms well and stable with farther forecast. However, the result of new device allocation recommendation doesn’t meet our expectation.

摘要 I
ABSTRACT. III
誌謝 V
List of Tables VIII
List of Figures IX
Chapter 1. Introduction 1
Chapter 2. Background and Related Work 4
2.1 Air Quality Analysis 4
2.2 Air Quality Prediction 6
Chapter 3. Problem Statement 8
Chapter 4. Proposed Method 10
4.1 System Framework 10
4.2 Utilized Features 12
4.2.1 Temporal Features 14
4.2.2 Spatial Features 14
4.3 Model of Methodology to Forecast PM2.5 Value 15
4.3.1 Temporal Forecast 17
4.3.2 Spatial Forecast 19
4.3.3 Smooth Strategy 21
4.4 New Devices Allocation Recommendation 24
Chapter 5. Experimental Evaluations 26
5.1. Experimental Data and Settings 26
5.1.1. Data Pre-Processing 29
5.1.2. Experimental Method 31
5.2. Air Quality Forecast 32
5.2.1 Experiment of Temporal Method 32
5.2.2 Experiment of Spatial Method 36
5.2.3 Experiment of Smooth Method 38
5.2.4 Performance of the Methodology 38
5.3. New Device Allocation Recommendation 39
Chapter 6. Conclusions and Future Work 40
6.1 Conclusions 40
6.2 Future Work 41
REFERENCES 42

[1] Aouizerats, B., Van Der Werf, G. R., Balasubramanian, R., ＆ Betha, R. (2015). Importance of transboundary transport of biomass burning emissions to regional air quality in Southeast Asia during a high fire event. Atmospheric chemistry and Physics, 15(1), 363-373.
[2] Baralis, E., Cerquitelli, T., Chiusano, S., Garza, P., ＆ Kavoosifar, M. R. (2016). Analyzing air pollution on the urban environment. In Information and Communication Technology, Electronics and Microelectronics (MIPRO), 2016 39th International Convention, 1464-1469.
[3] Bouarar, I., Brasseur, G., Granier, C., Petersen, K., Wang, X., Fan, Q., ... ＆ Mijling, B. (2016). Monitoring and Forecasting Air Quality over China: Results from the PANDA Modeling System. IGAC 2016 Science Conference (International Global Atmospheric Chemistry).
[4] Cagliero, L., Cerquitelli, T., Chiusano, S., Garza, P., Ricupero, G., ＆ Xiao, X. (2016). Modeling correlations among air pollution-related data through generalized association rules. In Smart Computing (SMARTCOMP), 2016 IEEE International Conference, 1-6.
[5] Cagliero, L., Cerquitelli, T., Chiusano, S., Garza, P., ＆ Ricupero, G. (2016). Discovering air quality patterns in urban environments. In Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing: Adjunct, 25-28.
[6] Chen, J., Chen, H., Zheng, G., Pan, J. Z., Wu, H., ＆ Zhang, N. (2014). Big smog meets web science: smog disaster analysis based on social media and device data on the web. Proceedings of the 23rd international conference on world wide web, 505-510.
[7] Crippa, M., Janssens-Maenhout, G., Dentener, F., Guizzardi, D., Sindelarova, K., Muntean, M., ... ＆ Granier, C. (2016). Forty years of improvements in European air quality: regional policy-industry interactions with global impacts. Atmospheric Chemistry and Physics, 16(6), 3825-3841.
[8] Cuchiara, G. C., Rappenglück, B., Rubio, M. A., Lissi, E., Gramsch, E., ＆ Garreaud, R. D. (2017). Modeling study of biomass burning plumes and their impact on urban air quality; a case study of Santiago de Chile. Atmospheric Environment, 166, 79-91.
[9] Domańska, D., ＆ Łukasik, S. (2016). Handling high-dimensional data in air pollution forecasting tasks. Ecological Informatics, 34, 70-91.
[10] Dong, Y., Wang, H., Zhang, L., ＆ Zhang, K. (2016). An improved model for PM2.5 inference based on support vector machine. Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD), 2016 17th IEEE/ACIS International Conference, 27-31.
[11] Gromke, C., Jamarkattel, N., ＆ Ruck, B. (2016). Influence of roadside hedgerows on air quality in urban street canyons. Atmospheric Environment, 139, 75-86.
[12] Hsieh, H. P., Lin, S. D., ＆ Zheng, Y. (2015). Inferring air quality for station location recommendation based on urban big data. Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 437-446.
[13] Jiang, W., Wang, Y., Tsou, M. H., ＆ Fu, X. (2015). Using social media to detect outdoor air pollution and monitor air quality index (AQI): a geo-targeted spatiotemporal analysis framework with Sina Weibo (Chinese Twitter). PloS one, 10(10), e0141185.
[14] Lee, Y. H., Shindell, D. T., Faluvegi, G., ＆ Pinder, R. W. (2015). Potential impact of a US climate policy and air quality regulations on future air quality and climate change. Atmospheric Chemistry ＆ Physics Discussions, 15(21).
[15] Li, Y., Huang, J., ＆ Luo, J. (2015). Using user generated online photos to estimate and monitor air pollution in major cities. Proceedings of the 7th International Conference on Internet Multimedia Computing and Service, 79.
[16] Lu, X., Wang, Y., Huang, L., Yang, W., ＆ Shen, Y. (2016). Temporal-Spatial Aggregated Urban Air Quality Inference with Heterogeneous Big Data. International Conference on Wireless Algorithms, Systems, and Applications, 414-426.
[17] Lv, B., Zhang, B., ＆ Bai, Y. (2016). A systematic analysis of PM2. 5 in Beijing and its sources from 2000 to 2012. Atmospheric Environment, 124, pp. 98-108.
[18] Martins, V., Moreno, T., Mendes, L., Eleftheriadis, K., Diapouli, E., Alves, C. A., ... ＆ Minguillón, M. C. (2016). Factors controlling air quality in different European subway systems. Environmental research, 146, 35-46.
[19] Mei, S., Li, H., Fan, J., Zhu, X., ＆ Dyer, C. R. (2014). Inferring air pollution by sniffing social media. In Proceedings of the 2014 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, 534-539.
[20] Millstein, D., Wiser, R., Bolinger, M., ＆ Barbose, G. (2017). The climate and air-quality benefits of wind and solar power in the United States. Nature Energy, 2(9), 17134.
[21] Shirmohammadi, F., Sowlat, M. H., Hasheminassab, S., Saffari, A., Ban-Weiss, G., ＆ Sioutas, C. (2017). Emission rates of particle number, mass and black carbon by the Los Angeles International Airport (LAX) and its impact on air quality in Los Angeles. Atmospheric Environment, 151, 82-93.
[22] Wang, S., Paul, M. J., ＆ Dredze, M. (2015). Social media as a sensor of air quality and public response in China. Journal of medical Internet research, 17(3).
[23] Zheng, Y., Liu, F., ＆ Hsieh, H. P. (2013). U-Air: When urban air quality inference meets big data. Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining,1436-1444.
[24] Zheng, Y., Chen, X., Jin, Q., Chen, Y., Qu, X., Liu, X., ... ＆ Sun, W. (2014). A cloud-based knowledge discovery system for monitoring fine-grained air quality. preparation, Microsoft Tech Report, http://research. microsoft. com/apps/pubs/default.
[25] Zhang, Y., Ding, A., Mao, H., Nie, W., Zhou, D., Liu, L., ... ＆ Fu, C. (2016). Impact of synoptic weather patterns and inter-decadal climate variability on air quality in the North China Plain during 1980–2013. Atmospheric Environment, 124, 119-128.
[26] Zheng, Y., Yi, X., Li, M., Li, R., Shan, Z., Chang, E., ＆ Li, T. (2015). Forecasting fine-grained air quality based on big data. Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2267-2276.
[27] Zhuang, Y., Lin, F., Yoo, E. H., ＆ Xu, W. (2015). Airsense: A portable context-sensing device for personal air quality monitoring. Proceedings of the 2015 Workshop on Pervasive Wireless Healthcare, 17-22.
[28] Zhu, J. Y., Zhang, C., Zhang, H., Zhi, S., Li, V. O., Han, J., ＆ Zheng, Y. (2017). pg-Causality: Identifying Spatiotemporal Causal Pathways for Air Pollutants with Urban Big Data. IEEE Transactions on Big Data.