類神經網路結合時間-面積法於降雨-逕流模式之研究

Integration of Artificial Neural Networks with Time-Area Method in Rainfall-Runoff Modeling

張倍源(Pei-Yuan Chang)

指導教授 ： 張麗秋

淡江大學/工學院/水資源及環境工程學系碩士班/碩士(2021年)

https://doi.org/10.6846/TKU.2021.00313

摘要

近年來因為全球暖化導致全球變遷，世界各地極端事件發生頻繁，臺灣也受到氣候變遷影響，颱風愈趨於強烈，強烈颱風出現次數逐漸上升，還有降雨時間、強度有改變，如降雨天數變少，總雨量不變，意即短延時強降雨事件次數增加，對臺灣各地可能造成災害，對水庫操作也是一種挑戰。
本研究以石門水庫集水區為研究區域，主要目的為探討空間分布差異與集流時間對降雨-逕流模式之影響，透過倒傳遞類神經網路以不同分區方式建置模型進行未來1-3小時入流量預測。本研究共建置三種模式，模式一以雨量站資料直接作為輸入因子進行模式訓練；模式二以常見的子集水區分區方法將石門水庫集水區進行分區，再將QPESUMS網格雨量資料進行分割進行訓練；模式三結合時間-面積法並考慮降雨時空分布特性進行分區進行訓練。為比較模式優劣，以RMSE、R^2、洪峰流量差、洪峰時間差等四種評估指標進行分析。
結果顯示將所蒐集到的QPESUMS資料繪製空間分布圖加以觀察，石門水庫集水區在季風型豪大雨事件之降雨時空分布受到西南季風或東北季風影響走向不同，降雨空間分布隨著降雨稽延呈現由東往西移動；為能更精準掌握集水區降雨量與分區集水時間對逕流量預報之影響，本研究探討兩種分區方法計算分需降雨量，一為常見的子集水區分區方法、另一為結合時間-面積法並考慮降雨時空分布特性進行分區，結果以後者有較好的表現；三種模式以評估指標分析比較後，以模式三有最佳結果，可推論降雨-逕流模式會受到空間分布差異與集流時間之影響。

關鍵字

類神經網路 ； 倒傳遞類神經網路 ； 降雨-逕流 ； 颱風 ； 豪雨 ； 時間-面積法

並列摘要

In recent years, because global warming has led to global changes, extreme events have occurred frequently all over the world, and Taiwan has also been affected by climate change. Typhoons are becoming more and more intense, the occurrence times of strong typhoons are gradually increasing, and the rainfall time and intensity have changed. If the number of rainfall days decreases, the total rainfall remains unchanged, that is, the number of short delay heavy rainfall events increases, which may cause disasters to all parts of Taiwan, It is also a challenge for reservoir operation.
Taking the catchment area of Shimen Reservoir as the research area, the main purpose of this study is to explore the effects of spatial distribution differences and concentration time on rainfall runoff model, and build models in different zoning methods through back-propagation neural network to predict the inflow in the next 1-3 hours. There are three models in this study. The first model takes the rainfall station data as the input factor for model training; In model 2, the catchment area of Shimen Reservoir is partitioned by the common subset water area zoning method, and then the QPESUMS grid rainfall data is segmented for training; Model 3 combines the time area method and considers the temporal and spatial distribution characteristics of rainfall for zoning training. In order to compare the advantages and disadvantages of the model, four evaluation indexes such as RMSE, R^2, peak discharge difference and peak time difference are analyzed.
The results show that the spatial distribution of rainfall in the catchment area of Shimen Reservoir is affected by the southwest monsoon or northeast monsoon, and the spatial distribution of rainfall moves from east to west with the extension of rainfall; In order to more accurately grasp the impact of catchment rainfall and regional catchment time on runoff forecasting, this study discusses two zoning methods to calculate the divided rainfall demand, one is the common subset water area zoning method, the other is the zoning method combined with the time-area method and considering the temporal and spatial distribution characteristics of rainfall. The results show that the latter has a better performance; After analyzing and comparing the three models with evaluation indexes, model 3 has the best results. It can be inferred that the rainfall runoff model will be affected by the difference of spatial distribution and concentration time.

並列關鍵字

Artificial Neural Networks ； Back Propagation Neural Network(BPNN) ； Rainfall-Runoff ； Typhoon, Extremely heavy rain ； Time-Area curve

參考文獻

1.Chen, S. H., Lin, Y. H., Chang, L. C., & Chang, F. J., 2006. The strategy of building a flood forecast model by neuro‐fuzzy network. Hydrological Processes: An International Journal, 20(7), 1525-1540.

2.Chen, Y.H., Chang, F.J., 2009. Evolutionary artificial neural networks for hydrological systems forecasting. Journal of Hydrology 367, 125-137.

3.Chen, N., Xiong, C., Du, W., Wang, C., Lin, X., & Chen, Z., 2019. An improved genetic algorithm coupling a back-propagation neural network model (IGA-BPNN) for water-level predictions. Water, 11(9), 1795.

4.Chiang, Y. M., Chang, L. C., & Chang, F. J., 2004. Comparison of static-feedforward and dynamic-feedback neural networks for rainfall–runoff modeling. Journal of hydrology, 290(3-4), 297-311.

5.Dai, H., & MacBeth, C., 1997. Effects of learning parameters on learning procedure and performance of a BPNN. Neural networks, 10(8), 1505-1521.