臺灣博碩士論文加值系統

在已開發國家中或是開發中國家，空氣污染物對於社會大眾的健康影響都是一致的。在空
氣微粒污染物中PM2.5是懸浮微粒(particulate matter，指空氣中極小的懸浮微粒)，由於尚未有
對於懸浮微粒濃度門檻值對於人體的影響，且懸浮微粒濃度對於擁有呼吸道系統疾病的人體影
響也有所不同，因此任何的標準或是規定皆無法完全的保護社會大眾。因此預測PM2.5在未來
的數值，是很重要的議題，透過預測可以在浮微粒濃度升高前，提醒社會大眾對於懸浮微粒的
事先的防護與預防。
本文以ETL(Extract-Transform-Load)的框架整理了2013~2017年環保署與氣象局提供的歷史
數據，透過AKIMA將缺失值補齊，透過整理將資料分類成三種不同污染源的資料集，利用
Aggregation Model 的方式，分別使用三種不同污染源的時間序列資料集，建立三個LSTM (
Long Short-Term Memory networks)子神經網路，以得到在三種不同污染源的資料中的預測特徵
資料，透過融合層結合三個子神經網路產出的預測特徵，並輸出到全連接層中透過反向傳播分
別給予其隱藏層不同的權重，最後得到未來1到8個小時預測的PM2.5數值。在與現有的方法
ANN與LSTM比較後，第一個小時的準確率在RMSE上比LSTM減少0.15誤差、在MAE誤差減
少0.11，與ANN比較後在RMSE誤差減少0.75、在MAE誤差減少0.54。

In developed countries or developing countries, the effects of air pollutants on the health of the
public are consistent. PM2.5 is a suspended particle In the airborne particulate pollutants. There is no impact on the human body for the concentration threshold of suspended particulates. The concentration of suspended particulates is for humans with diseases of the respiratory system. The impact is also different, so no standard or regulation can completely protect the public. Therefore, predicting the value of PM2.5 in the future is an important issue.
This paper uses data provided by the Environmental Protection Agency and Central Weather Bureau from 2013 to 2017. Dividing a data set into data sets of three different sources of pollution. The Aggregation Model uses three types time series data sets of different pollution sources to establish three LSTM (Long Short-Term Memory networks) sub-neural networks to obtain prediction characteristics in three different pollution sources. Combining the predicted features produced by the three sub-neural networks, and outputting prediction feature to the fully connected layer, respectively, giving different weights by the hidden layers from back propagation, and finally obtaining the PM2.5 values predicted in the next 1 to 8 hours. After comparing with the existing methods ANN and LSTM, the accuracy of the next hour is 0.15 less than the LSTM and 0.11 in the MAE. The RMSE is reduced by 0.75 and the MAE error is reduced by 0.54 after comparison with the ANN.

ABSTRACT II
圖目錄 VI
表目錄 VII
第一章 緒論 1
1.1 現況與問題 1
1.2 目標與貢獻 2
1.3 論文架構 3
第二章 背景與相關研究 5
2.1 背景 5
2.1.1 Tensorflow 5
2.1.2 Keras 6
2.1.3 MongoDB 7
2.2 神經網路 7
2.2.1 RNN 8
2.2.2 LSTM 9
2.2.3 Dense and Dropout 13
2.3 相關研究 14
第三章 資料處理架構 16
3.1 資料處理平台 16
3.1.1 資料來源 16
3.1.2 Extract-Transform-Load 16
3.2 資料預處理 18
3.2.1 資料對齊 18
3.2.2 資料補值 23
3.2.3 資料標準化 26
第四章 模型設計與實驗架構 28
4.1 Aggregation Model 28
4.2 Aggregation with LSTM 29
4.2.1 模型架構 29
4.2.2 模型建立 40
第五章 實驗結果與比較 45
5.1 實驗環境 45
5.2 評估方式 45
5.2.1 RMSE 45
5.2.2 MAE 46
5.2.3 MAPE 46
5.3 預測結果 46
5.3.1 預測結果分析比較 49
5.3.2 準確度分析 53
第六章 結論與未來目標 60
參考文獻 62
附錄 65

[1] D. W. Dockery, et al. “An association between air pollution and mortality in six
U.S. cities,” New England Journal of Medicine, vol. 329, no. 24, pp. 1753-1759,
1993.
[2] C. A. Pope, et al. “Cancer, cardiopulmonary mortality, and long-term exposure to
fine particulate air pollution,” J. Amer. Med. Assoc., vol. 287, no. 9, pp. 1132-
1141, 2002.
[3] C. Zhu. China tackles the health effects of air pollution. Lancet 382.9909, Dec,
2013.
[4] World Health Organization, “WHO AIR QUALITY GUIDELINES GLOBAL
UPDATE 2005,” Report on a working group meeting, Bonn, Germany, 2005.
[5] Q. Zhou, H. Jiang, J. Wang, & J. Zhou, “A hybrid model for PM2.5 forecasting
based on ensemble empirical mode decomposition and a general regression
neural network,” Science of the Total Environment, pp. 264-274, 2014.
[6] M. A. Elangasinghe, N. Singhal, & K. N. Dirks, “Complex time series analysis
of PM10 and PM2.5 for a coastal site using artificial neural network modelling
and k-means clustering,” Atmospheric Environment, pp. 106-116, 2014.
[7] X. Hu, A. Lance, & L. A. Waller, “Estimating ground-level PM2.5
concentrations in the Southeastern United States using MAIAC AOD retrievals
and a two-stage model,” Remote Sensing of Environment, pp. 220-232, 2014.
[8] F. Deng, L. Ma, X. Gao, & J. Chen, “The MR-CA Models for Analysis of
Pollution Sources and Prediction of PM2.5” IEEE Transactions on Systems,
Man, and Cybernetics: Systems, 2017.
[9] L. Breiman, “Bagging predictors,” Machine Learning, vol. 24, no. 2, pp. 123-
140, 1996.
[10] S. Hochreiter, and J. Schmidhuber, “Long Short-Term Memory,” Neural
Computation. vol. 9 ,no. 8 , pp. 1735-1780, 1997.
[11] M. Schuster, and K. K. Paliwal, “Bidirectional recurrent neural networks,” IEEE
Transactions on Signal Processing, vol. 45, no. 11, pp. 2673-2681, 1997.
[12] Y. Bengio, P. Simard, & P. Frasconi “Learning long-term dependencies with
gradient descent is difficult,” IEEE Trans Neural Netw, vol. 5, no. 2, pp. 157-66,
1994
[13] C. A. Pope, M. Thun, M. Namboodiri, D. Dockery, J. Evans, F. Speizer, , and C.

66

Heath, “Particulate air pollution as a predictor of mortality in a prospective study
of U.S. adults.” American Journal Respiratory Critical Care Medicine, vol. 151,
no.3, pp. 669-674, 1995.
[14] J. B. Ordieresa, E. P. Vergara, R.S. Capuz, “Neural network prediction model for
fine particulate matter (PM2.5) on the USeMexico border in El Paso (Texas) and
Ciudad Juarez (Chihuahua),” Environmental Modelling & Software, pp. 547-
559, 2005.
[15] S. V. Barai, A. K. Dikshit, & S. Sharma, ”Neural Network Models for Air
Quality Prediction: A Comparative Study,” Soft Computing in Industrial
Applications, pp. 290-305, 2007.
[16] G. Corani, “Air quality prediction in Milan: feed-forward neural networks,
pruned neural networks and lazy learning,” Ecological Modelling, pp. 513-529,
2005.
[17] A. Russo, F. Raischel, & P. G. Lind, “Air quality prediction using optimal neural
networks with stochastic variables,” Atmospheric Environment, pp. 822-830,
2013.
[18] F. Mingjian, Z. Guocheng, Z. Xuxu, & Y. Zhongyi, “Study on Air Fine Particles
Pollution Prediction of Main Traffic Route Using Artificial Neural Network”
2011 International Conference on Computer Distributed Control and Intelligent
Environmental Monitoring, pp. 19-20, 2011.
[19] H. Akima, “A New Method of Interpolation and Smooth Curve Fitting Based on
Local Procedures,” Journal of the ACM, vol. 17, no. 4, pp. 589-602, 1970.
[20] D. E. Rumelhart, G. E. Hinton, & R. J. Williams, “Learning representations by
back-propagating errors,” Nature, vol. 323, no. 6088, pp. 533-536, 1986.
[21] F. J. Pineda, “Generalization of back propagation to recurrent networks,” Phys.
Rev. Lett, vol. 59, no. 19, pp. 2229-2232, 1988.
[22] Y. T. Tsai, Y. R. Zeng, and Y. S. Chang, “Air pollution forecasting using RNN
with LSTM,” The 4th IEEE Intl Conf on Big Data Intelligence and Computing,
Athens, Greece, 2018 (in press).