臺灣博碩士論文加值系統

由於近年全球人口提升，工業化的需求不斷增加，空汙所造成的環境汙染也日劇嚴重。現今大部分之PM2.5預測皆使用過去PM2.5濃度之歷史數據或是使用相關空氣汙染因子及PM2.5濃度作為訓練數據對空氣品質做預測。2020年中華民國行政院環境保護署空氣品質監測網所提供的監測資料中，全台灣共有217個監測點，這些監測站並無平均分布於台灣各區域，而是分布於工業密集的地方，並且不是全部的監測站都有監測PM2.5數值。因此，如何使用空汙因子如SO2、NO2與PM10準確地預測出PM2.5懸浮微粒，成為現今非常重要的研究議題。本論文將SO2、NO2及PM10前三分之二筆數據做為LSTM模型與分類器模型的訓練數據，再將後三分之一筆數據做為此兩種模型的測試數據。LSTM模型預測每小時之SO2、NO2及PM10三項空汙相關因子，最後將預測後的三項空汙因子數據輸入分類器中預測PM2.5濃度等級。PM2.5濃度等級劃分參考環保署標準將PM2.5濃度等級區分為良好/普通、不健康、非常不健康與危害等四個等級。比較六種LSTM模型，以均方根誤差(RMSE)來評估模型的精確度。此外，為了評估模型的訓練成本，我們比較其模型參數(Params)，找出預測SO2、NO2及PM10使用參數量最少的LSTM模型。最後比較五種不同的分類器，找出預測PM2.5濃度等級準確率最高的模型，達到使用空汙因子濃度數據預測PM2.5濃度等級的目標。實驗結果顯示六種LSTM模型預測SO2、NO2及PM10濃度時均方根誤差皆相當，而使用標準LSTM模型預測SO2有最少的模型參數，使用堆疊LSTM模型預測NO2與PM10有最少模型參數，結合K-近鄰分類器會得到最高的準確率83.45%。

As the global population has increased in recent years, the demand for industrialization has continued to increase, and the environmental pollution caused by air pollution has become increasingly serious. Most of the current PM2.5 forecasts use historical PM2.5 concentration data as training data or use relevant air pollution factors and PM2.5 concentrations to predict air quality. According to the monitoring data provided by Air Quality Monitoring Network of the Environmental Protection Agency of the Executive Yuan of Republic of China in 2020 , there are 217 monitoring points in Taiwan . These monitor stations are not evenly distributed over all regions of Taiwan , but are located in densely industrialized areas , and not all monitoring stations have PM2.5 values . Therefore, how to use air pollution factors such as SO2, NO2 and PM10 to accurately predict PM2.5 suspended particles has become a very important research topic today. This paper uses the LSTM deep learning model to use the first two-thirds of the data of SO2, NO2 and PM10 as training data, and then use the last third of the data as test data to predict the hourly rate. SO2, NO2 and PM10 three air pollution related factors. At the same time, the concentration data of SO2, NO2 and PM10 are input into the classifier for training, and then the three data of SO2, NO2 and PM10 predicted by the LSTM model are input into the classifier to predict PM2 .5 Concentration levels. Refer to the Environmental Protection Agency standards to classify PM2.5 concentration levels into four levels: good/normal, unhealthy, very unhealthy and hazardous. We need to find the method with the highest efficiency and accuracy to predict the PM2.5 concentration level, compare six LSTM models, and use the root mean square error (RMSE) to evaluate the accuracy of the model. In addition , in order to evaluate the training cost of the model , We compare its model parameters (Params) to find the LSTM model with the least amount of parameters for predicting SO2, NO2 and PM10. Finally, compare five different classifiers to find the model with the highest accuracy in predicting PM2.5 concentration levels, and achieve the goal of using air pollution factor concentration data to predict PM2.5 concentration levels. The experimental results show that the root mean square error of the six LSTM models predicting SO2, NO2 and PM10 concentrations are all equivalent, while using the standard LSTM model predicting SO2 has the least model parameters, using the stacked LSTM model to predict NO2 and PM10 have the least model parameters, combined with the K-nearest neighbor classifier will get the highest accuracy rate of 83.45%.

中文摘要.............................................. i
英文摘要 ............................................. ii
誌謝 ............................................. iii
目錄 ............................................. iv
表目錄 ............................................. v
圖目錄 ............................................. vi
第一章 緒論.......................................... 1
第二章 長短期記憶模型................................ 4
2.1.1長短期記憶..................................... 4
2.1.2雙向長短期記憶.................................. 8
2.2.1堆疊長短期記憶................................. 10
2.2.2巢狀長短期記憶................................. 12
第三章 長短期記憶結合機器學習分類器模型................ 15
3.1 KNN分類器..................................... 15
3.2 Bagging分類器................................. 17
3.3 Decision Tree分類器........................... 19
3.4 GBDT分類器...................................... 21
3.5.1 樸素貝葉斯分類器................................. 23
3.5.2 高斯樸素貝葉斯分類器............................ 24
3.5.3 多項式樸素貝葉斯分類器............................25
3.5.4 伯努力樸素貝葉斯分類器........................... 26
3.6 長短期記憶結合分類器機器學習方法...................27
第四章 空氣品質數據集之PM2.5濃度分類預測.................. 28
4.1 SO2預測結果分析............................ 30
4.2 NO2預測結果分析..................................34
4.3 PM10預測結果分析.................................38
4.4 PM2.5分類預測結果分析.............................43
第五章 結論.............................................46
參考文獻............................................. 47
Extended............................................. 49
Abstract............................................. 49

[1] A. Graves, M. Liwicki, S. Fernández, R. Bertolami, H. Bunke and J. Schmidhuber, "A Novel Connectionist System for Unconstrained Handwriting Recognition," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 31, no. 5, pp. 855-868, May 2009, doi: 10.1109/TPAMI.2008.137.
[2] Yuanyuan Wang, et al., 2017, "Water Quality Prediction Method Based on LSTM Neural Network", 2017 12th International Conference on Intelligent Systems and Knowledge Engineering (ISKE), pp. 24-26, November.
[3] Songyang Zhang, et al., 2018, "Fusing Geometric Features for Skeleton-Based Action Recognition Using Multilayer LSTM Networks", IEEE Transactions on Multimedia, Volume: 20, Issue:9, pp. 2330-2343, September.
[4] Khaled A. Althelaya, et al., 2018, "Evaluation of Bidirectional LSTM for Short- and Long-Term Stock Market Prediction", 2018 9th International Conference on Information and Communication Systems (ICICS), pp. 151-156, April.
[5] Joel Ruben Antony Moniz, David Krueger, 2017, "Nested LSTMs", Proceedings of the Ninth Asian Conference on Machine Learning, PMLR 77, pp. 530-544, January
[6] J. Zhou, Y. Lu, et al., "Sentiment Analysis of Chinese Microblog Based on Stacked Bidirectional LSTM", IEEE Access, vol. 7, pp. 38856-38866, 2019.
[7] T. Cover and P. Hart, "Nearest neighbor pattern classification," in IEEE Transactions on Information Theory, vol. 13, no. 1, pp. 21-27, January 1967, doi: 10.1109/TIT.1967.1053964.
[8] Breiman, L., Bagging Predictors, Machine Learning, 24(2), 1996.
[9]J.R. Quinlan.,1986, “Induction of Decision Trees”, Machine Learning,1,81-106
[10]決策樹（Decision Tree）簡介
https://blog.csdn.net/fengbingchun/article/details/78880934?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522159393659419725222420772%2522%252C%2522scm%2522%253A%252220140713.130102334..%2522%257D&request_id=159393659419725222420772&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~first_rank_ecpm_v3~pc_rank_v4-7-78880934.first_rank_ecpm_v3_pc_rank_v4&utm_term=Decision+Tree
[11]Gradient Boosting Decision Tree Algorithm Explained
https://towardsdatascience.com/machine-learning-part-18-boosting-algorithms-gradient-boosting-in-python-ef5ae6965be4
[12] Y. Ji, S. Yu and Y. Zhang, "A novel Naive Bayes model: Packaged Hidden Naive Bayes," 2011 6th IEEE Joint International Information Technology and Artificial Intelligence Conference, Chongqing, 2011, pp. 484-487, doi: 10.1109/ITAIC.2011.6030379.
[13] A. H. Jahromi and M. Taheri, "A non-parametric mixture of Gaussian naive Bayes classifiers based on local independent features," 2017 Artificial Intelligence and Signal Processing Conference (AISP), Shiraz, 2017, pp. 209-212, doi: 10.1109/AISP.2017.8324083.
[14]G. Singh, B. Kumar, L. Gaur and A. Tyagi, "Comparison between Multinomial and Bernoulli Naïve Bayes for Text Classification," 2019 International Conference on Automation, Computational and Technology Management (ICACTM), London, United Kingdom, 2019, pp. 593-596, doi: 10.1109/ICACTM.2019.8776800.
[15] 戴君瑋，2019，基於雙向巢狀長短期記憶之支持向量機分類器，雲林縣：虎尾科技大學， 7月
[16]Beijing Multi-Site Air-Quality Data Data Set
https://archive.ics.uci.edu/ml/datasets/Beijing+Multi-Site+Air-Quality+Data
[17] 林澔等，2019，基於深度學習技術應用於空氣品質 PM2.5預測
https://tanet2019.nsysu.edu.tw/assets/TANET2019_thesis/B2_009.pdf
[18] 基於隨機森林算法的PM2.5預測
https://zhuanlan.zhihu.com/p/83220850
[19]https://aihub.org.tw/platform/algorithm/search?q=[%E7%A9%BA%E6%B1%A1]
[20] R. D. Brook, S. Rajagopalan, C. A. Pope, J. R. Brook, A. Bhatnagar, A. V. Diez-Roux, et al., "Particulate matter air pollution and cardiovascular disease: An update to the scientific statement from the american heart association", Circulation, vol. 121, pp. 2331-2378, Jun. 2010.
[21] Y. Jiao, Z. Wang and Y. Zhang, "Prediction of Air Quality Index Based on LSTM," 2019 IEEE 8th Joint International Information Technology and Artificial Intelligence Conference (ITAIC), Chongqing, China, 2019, pp. 17-20, doi: 10.1109/ITAIC.2019.8785602.
[22]中華民國行政院環境保護署空氣品質監測網
https://airtw.epa.gov.tw/