近年來受到氣候變遷與全球暖化之影響，世界各地發生極端水文事件的頻率增加，對人類生命與財產安全造成極大威脅。臺灣降雨量在空間與時間分布呈現不均勻現象在近年來更趨嚴重，除了受到梅雨鋒面與颱風影響外，各縣市也因短延時強降雨頻繁發生，導致雨水下水道水位驟升溢出路面，短時間內即造成市區局部區域發生積淹水情形，甚至大範圍嚴重淹水災情，救災單位整備與應變時間相對變短。
    本研究以臺北市中山抽水站集水區為研究區域，應用長短期記憶模型(LSTM)與倒傳遞類神經網路(BPNN)建置人工智慧下水道水位預報模式，預報中山抽水站內池水位未來10至60分鐘之水位；依照不同輸入因子數量將模式分為兩種子模式，輸入I型（所有雨量站）與輸入II型（部分雨量站），針對雨量站間之距離調整輸入因子，分析兩種不同輸入因子對於LSTM與BPNN模式預測之準確性，並討論降雨與下水道水位變化之關係。
    由輸入I型與輸入II型綜合比較結果可得知，調整輸入項之雨量站對於模式訓練上有一定影響，顯示雨量站、下水道水位測站與抽水站有相當關聯性；LSTM模式在輸入I型與輸入II型之結果表現均優於BPNN模式，可證明LSTM模式在學習資料特性上有掌握水位關係，對於預測抽水站內池水位未來10至60分鐘有較佳之預測結果，有助於輔助抽水站操作人員在暴雨時期參考並做出即時且準確之決策，亦可作為抽水機啟閉之依據，漸少專業人員在判斷上的壓力。

In recent years, the occurrence of extreme hydrological events is more frequent, due to climate change and global warming, posing a great threat to human life and property security. The uneven distribution of rainfall in space and time in Taiwan has become more serious in recent years. In addition to the impact of plum rain(mei-yu) and typhoon, short duration high intensity storms also could flood counties and cities, the water overflowed from the sewer on to the road surface in a short period of time, emergency response agencies’ response time is insufficient.
    This study took the Zhongshan Pumping Station watershed in Taipei City as the research area. The artificial intelligence sewer water level forecasting model was constructed using the long-short term memory model (LSTM) and back-propagation neural network (BPNN), forecasting the water levels of the front storage pool in Zhongshan pumping station for the next 10 to 60 minutes (T+1~T+6). The model is divided into two sub-models, the first submodel uses all rain stations as inputs (type I) and second submodel selects partial rain station (type II). This study analyzed the forecast result of LSTM and BPNN models with two different input factors, and discussed the relationship between rainfall and sewer water level. 
    In comparison with the results of type I, type II showed that there is a considerable correlation between rain station, sewer station and pumping station, adjusting model input has an effect on model training. In conclusion, this study proved that the LSTM model has get the hang of data property in training model, the results also demonstrate that the LSTM model is more accurately forecast water levels of the front storage pool in type I and type II than BPNN model. The proposed methodology can provide water level information to decision-makers and residents for taking precautionary measures against flooding.

謝誌I
中文摘要III
AbstractV
目錄VII
圖目錄IX
表目錄XII
第一章	前言1
1.1	研究緣起1
1.2	研究目的與方法2
1.3	論文架構3
第二章	文獻回顧4
2.1	運用時間序列分析方法於水文預測之發展4
2.2	長短期記憶模型之應用6
2.3	倒傳遞類神經網路之應用8
第三章	理論概述10
3.1	長短期記憶模型10
3.1.1	長短期記憶模型架構11
3.1.2	長短期記憶模型演算法12
3.2	倒傳遞類神經網路14
3.2.1	誤差倒傳遞演算法15
3.3	參數設定18
3.3.1	活化函數18
3.3.2	最佳化演算法23
3.3.3	過度擬合25
第四章	研究案例28
4.1	研究區域28
4.2	資料蒐集32
4.3	模式架構36
4.3.1	資料前處理36
4.3.2	模式建置39
4.4	評估指標42
第五章	結果與討論43
5.1	結果分析43
5.2	綜合討論80
第六章	結論與建議83
6.1	結論83
6.2	建議85
參考文獻	86
圖目錄
圖3.1 長短期記憶模型(LSTM)架構圖12
圖3.2 倒傳遞類神經網路(BPNN)架構圖14
圖3.3 Sigmoid函數19
圖3.4 Tanh函數20
圖3.5 ReLU函數21
圖3.6 過度擬合示意圖25
圖3.7 使用Dropout前後之網路架構圖27
圖4.1 臺北市下水道管線圖29
圖4.2 研究區域之抽水站、雨水下水道水位測站與雨量站分布圖30
圖4.3 臺北市集水區2019年07月22日累積日雨量35
圖4.4 原始觀測資料異常（以內池水位資料為例）37
圖4.5 觀測資料除錯示意圖37
圖4.6 U0043下水道水位測站異常情形38
圖4.7 模式建置流程架構圖41
圖5.1 輸入I型LSTM模式T+1抽水站內池水位預測與觀測比較圖50
圖5.2 輸入I型LSTM模式T+2抽水站內池水位預測與觀測比較圖51
圖5.3 輸入I型LSTM模式T+3抽水站內池水位預測與觀測比較圖52
圖5.4 輸入I型LSTM模式T+4抽水站內池水位預測與觀測比較圖53
圖5.5 輸入I型LSTM模式T+5抽水站內池水位預測與觀測比較圖54
圖5.6 輸入I型LSTM模式T+6抽水站內池水位預測與觀測比較圖55
圖5.7 輸入I型BPNN模式T+1抽水站內池水位預測與觀測比較圖56
圖5.8 輸入I型BPNN模式T+2抽水站內池水位預測與觀測比較圖57
圖5.9 輸入I型BPNN模式T+3抽水站內池水位預測與觀測比較圖58
圖5.10輸入I型BPNN模式T+4抽水站內池水位預測與觀測比較圖59
圖5.11輸入I型BPNN模式T+5抽水站內池水位預測與觀測比較圖60
圖5.12輸入I型BPNN模式T+6抽水站內池水位預測與觀測比較圖61
圖5.13輸入II型LSTM模式T+1抽水站內池水位預測與觀測比較圖68
圖5.14輸入II型LSTM模式T+2抽水站內池水位預測與觀測比較圖69
圖5.15輸入II型LSTM模式T+3抽水站內池水位預測與觀測比較圖70
圖5.16輸入II型LSTM模式T+4抽水站內池水位預測與觀測比較圖71
圖5.17輸入II型LSTM模式T+5抽水站內池水位預測與觀測比較圖72
圖5.18輸入II型LSTM模式T+6抽水站內池水位預測與觀測比較圖73
圖5.19輸入II型BPNN模式T+1抽水站內池水位預測與觀測比較圖74
圖5.20輸入II型BPNN模式T+2抽水站內池水位預測與觀測比較圖75
圖5.21輸入II型BPNN模式T+3抽水站內池水位預測與觀測比較圖76
圖5.22輸入II型BPNN模式T+4抽水站內池水位預測與觀測比較圖77
圖5.23輸入II型BPNN模式T+5抽水站內池水位預測與觀測比較圖78
圖5.24輸入II型BPNN模式T+6抽水站內池水位預測與觀測比較圖79

表目錄
表4-1  研究區之抽水站基本資料	30
表4-2  研究區之雨水下水道水位監測站基本資料表	31
表4-3  研究區之雨量站基本資料表	31
表4-4  歷年抽水站、水位站、雨量站資料統計概要表	32
表4-5  鄰近雨量站不同延時之最大降雨量與發生時刻	34
表4-6  模式訓練參數設定	40
表5-1  LSTM與BPNN模式使用場次	43
表5-2  模式輸入時刻測試組合	45
表5-3  輸入I型LSTM與BPNN模式各站延時設定表	47
表5-4  輸入I型LSTM模式三階段預測結果比較表	49
表5-5  輸入I型BPNN模式三階段預測結果比較表	49
表5-6  輸入II型LSTM與BPNN模式各站延時設定表	62
表5-7  不同雨量站輸入項測試組合	63
表5-8  不同輸入組合之結果表(以預測未來10分鐘為例)	64
表5-9  輸入II型LSTM模式三階段預測結果比較表	66
表5-10  輸入II型BPNN模式三階段預測結果比較表	66
表5-11  輸入I型與輸入II型之LSTM模式結果比較	81
表5-12  輸入I型與輸入II型之BPNN模式結果比較	81

1. Altché, F., Fortelle, A.d.L., (2017). An LSTM network for highway trajectory prediction, 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC), pp. 353-359.
2. Apaydin, H. et al., (2020). Comparative Analysis of Recurrent Neural Network Architectures for Reservoir Inflow Forecasting. Water, 12(5).
3. Benmahdjoub, K., Ameur, Z., Boulifa, M., (2013). Forecasting of Rainfall Using Time Delay Neural Network in Tizi-Ouzou (Algeria). Energy Procedia, 36: 1138-1146.
4. Bouktif, S., Fiaz, A., Ouni, A., Serhani, M.A., (2018). Optimal Deep Learning LSTM Model for Electric Load Forecasting using Feature Selection and Genetic Algorithm: Comparison with Machine Learning Approaches †. Energies, 11(7).
5. Chang, F.-J., Wang, Y.-C., Tsai, W.-P., (2016). Modelling Intelligent Water Resources Allocation for Multi-users. Water Resources Management, 30(4): 1395-1413.
6. Chang, F.J., Chang, L.-C., Huang, H.-L., (2002). Real-time recurrent learning neural network for stream-flow forecasting. Hydrological Processes, 16(13): 2577-2588.
7. Chang, L.-C., Shen, H.-Y., Chang, F.-J., (2014). Regional flood inundation nowcast using hybrid SOM and dynamic neural networks. Journal of Hydrology, 519: 476-489.87
8. Das, U.K., Roy, P., Ghose, D.K., (2019). Modeling water table depth using adaptive Neuro-Fuzzy Inference System. ISH Journal of Hydraulic Engineering, 25(3): 291-297.
9. Ghose, D.K., Samantaray, S., (2019). Integrated Sensor Networking for Estimating Ground Water Potential in Scanty Rainfall Region: Challenges and Evaluation. In: Mishra, B.B. et al. (Eds.), Computational Intelligence in Sensor Networks. Springer Berlin Heidelberg, Berlin, Heidelberg, pp. 335-352. 
10. Hameed, M. et al., (2017). Application of artificial intelligence (AI) techniques in water quality index prediction: a case study in tropical region, Malaysia. Neural Computing and Applications, 28(1): 893-905.
11. Hochreiter, S., Schmidhuber, J., (1997). Long Short-Term Memory. Neural Computation, 9(8): 1735-1780.
12. Huang, K. et al., (2020). Self-Learning Algorithm to Predict Indoor Temperature and Cooling Demand from Smart WiFi Thermostat in a Residential Building. Sustainability, 12(17): 7110.
13. Huang, X. et al., (2019). Groundwater Recharge Prediction Using Linear Regression, Multi-Layer Perception Network, and Deep Learning. Water, 11(9).
14. Kao, I.F., Zhou, Y., Chang, L.-C., Chang, F.-J., (2020). Exploring a Long Short-Term Memory based Encoder-Decoder framework for multi-step-ahead flood forecasting. Journal of Hydrology, 583: 124631.88
15. Liu, W.-C., Chung, C.-E., (2014). Enhancing the Predicting Accuracy of the Water Stage Using a Physical-Based Model and an Artificial Neural Network-Genetic Algorithm in a River System. Water, 6(6).
16. Moriya, Y., Jones, G., (2018). LSTM Language Model Adaptation with Images and Titles for Multimedia Automatic Speech Recognition. 2018 IEEE Spoken Language Technology Workshop (SLT): 219-226.
17. Park, J. et al., (2017). PM10 density forecast model using long short term memory, 2017 Ninth International Conference on Ubiquitous and Future Networks (ICUFN), pp. 576-581.
18. Raza, A. et al., (2019). Heartbeat Sound Signal Classification Using Deep Learning. Sensors, 19(21).
19. Samantaray, S., Sahoo, A., (2020). Appraisal of Runoff Through BPNN, RNN, and RBFN in Tentulikhunti Watershed: A Case Study, pp. 258-267. 
20. Shen, H.Y., Chang, L.C., (2013). Online multistep-ahead inundation depth forecasts by recurrent NARX networks. Hydrol. Earth Syst. Sci., 17(3): 935-945.
21. Siami-Namini, S., Tavakoli, N., Namin, A.S., (2018). A Comparison of ARIMA and LSTM in Forecasting Time Series, 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA), pp. 1394-1401.
22. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R., (2014). Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res., 15(1): 1929–1958.89
23. Suh, D., Kim, H., Kim, J., (2015). Estimation of Water Demand in Residential Building Using Machine Learning Approach, 2015 5th International Conference on IT Convergence and Security (ICITCS), pp. 1-2.
24. Yao, H., Zhang, X., Zhou, X., Liu, S., (2019). Parallel Structure Deep Neural Network Using CNN and RNN with an Attention Mechanism for Breast Cancer Histology Image Classification. Cancers, 11(12): 1901.
25. Zainorzuli, S.M., Abdullah, S.A.C., Adnan, R., Ruslan, F.A., (2019). Comparative Study of Elman Neural Network (ENN) and Neural Network Autoregressive With Exogenous Input (NARX) For Flood Forecasting, 2019 IEEE 9th Symposium on Computer Applications & Industrial Electronics (ISCAIE), pp. 11-15.
26. Zhang, J., Zhu, Y., Zhang, X., Ye, M., Yang, J., (2018). Developing a Long Short-Term Memory (LSTM) based model for predicting water table depth in agricultural areas. Journal of Hydrology, 561: 918-929.
27. Zhou, Y., Chang, F.-J., Chang, L.-C., Kao, I.F., Wang, Y.-S., (2019). Explore a deep learning multi-output neural network for regional multi-step-ahead air quality forecasts. Journal of Cleaner Production, 209: 134-145.
28. 呂英睿, (2014). 智慧型抽水站排水系統水位預報及操作策略整合
模式, 國立臺灣大學, 1-123 pp.
29. 張凱堯, (2009). 人工智慧於都市防洪排水系統控制之研究, 國立臺灣大學, 臺北市, 119 pp.90
30. 張斐章、張麗秋, (2015).「類神經網路導論-原理與應用第二
版」，滄海書局。