機器學習結合數值模式對集水區土砂收支預報之研究

Combination of Machine-learning Method and Sediment Transport Model to Forecast Sediment Budget in Watershed

張書豪(Shu-Hao Chang)

指導教授 ： 何昊哲

國立臺灣大學/工學院/土木工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202002800

摘要

台灣地理位置特殊，除位處颱風路徑上之外，更位於環太平洋地震帶，是複合型自然災害高危險區。過去集水區的土砂災害事都是單純以物理模式或是數值模式，研析各種情境之模擬結果，用於探討災害發生成因與減災之策略擬定，模擬結果與實際災害發生的位置與規模仍有差距。發展一個可靠的土砂收支預報模式，除有助於集水區的土砂管理外，更可以提早做出災害因應。
機器學習具有模擬非線性系統、運算快速與持續學習之優勢，已經被廣泛地應用於各水文領域中，包括流量預測，水位預測及泥砂濃度，皆可用機器學習進行分析、推估與預報。而數值模式的優勢則是可提供不同情境的模擬結果，並根據需求調整不同參數產生許多模擬資料。本研究以機器學習結合數值模式的方法，提出一個透過前端的水文資料及土壤崩塌沖蝕量來預估集水區的土砂流量趨勢之方法。
本研究以大漢溪流域為研究範圍，利用美國陸軍工兵團所發展之水文模式(HEC-HMS)進行降雨-逕流模擬產製流量歷線，再利用適用於石門水庫之土壤沖蝕、崩塌經驗公式產製崩塌量及沖蝕量並將其量值合理的轉化為具有時間關係的砂量歷線。，接著結合二維水理輸砂模式(SRH-2D)進行河道土砂運移模擬，產生不同情境資料提供機器學習(SVM、BPNN、RBFNN網路)進行學習，由模擬結果可知網路三者皆可掌握土砂趨勢，可提供未來1到6小時的即時砂量趨勢預報，對於未來不同颱洪事件下的土砂趨勢預測可提供一定的科學依據。

關鍵字

機器學習 ； 土砂收支 ； 類神經網路 ； SRH2D ； HEC-HMS

並列摘要

Taiwan's unique geographic location makes it a high-risk area for complex natural disasters. In the past, the catchment area has been used to simulate and analyze various scenarios using either physical or numerical models. The simulated results are used to investigate the causes of disasters and to formulate mitigation strategies. However, there are still gaps between simulation and reality. It is necessary to have a reliable prediction model for the revenue and expenditure of natural disasters in the catchment area.
Machine learning has the advantage of simulating non-linear systems, rapid extrapolation and continuous learning. It has been widely used in hydrological fields, including flow, water level and sediment concentration. Machine learning has been proven to be used for analysis, prediction and forecasting. The advantage of the numerical mode is that it can provide simulation results in different situations, and can be adjusted according to the needs of different parameters to generate many simulation data. The advantage of the numerical mode is that it can provide simulation results in different scenarios, and different parameters can be adjusted according to the need to generate a large number of simulation data. The study is based on a combination of machine learning and numerical models. In this study, we propose a method for predicting the sediment flow in a catchment using in-situ hydrological data by combining machine learning with numerical models.
In this study, the hydrological model developed by the U.S. Army Corps of Engineers (HEC-HMS) is used to determine the flow in the Dahan Creek watershed. The input conditions of the simulation as well as the amount of soil collapse and erosion are used as input conditions for the soil-sand simulation, and are combined with the hydraulic sand transport model developed by the U.S.B.R. (SRH-2D) to simulate the movement of sand and soil in the river channel, and generate simulation data under different scenarios to provide a machine learning model (BPNN、RBFNN、SVM). The simulation results show that all three of the network can grasp the trend of soil and sand, and can provide real-time sendiment flow trend forecasts in the next 1 to 6 hours, and provide a certain scientific basis for forecasting soil and sand trends under different typhoon events in the future.

並列關鍵字

Machine-learning ； Sediment budget ； ANNs ； SRH2D ； HEC-HMS

參考文獻

Afan, H. A., El-Shafie, A., Yaseen, Z. M., Hameed, M. M., Mohtar, W. H. M. W., Hussain, A. (2015). ANN based sediment prediction model utilizing different input scenarios. Water Resources Management, 29(4), 1231-1245.

Ajami, N. K., Duan, Q., Sorooshian, S. (2007). An integrated hydrologic Bayesian multimodel combination framework: Confronting input, parameter, and model structural uncertainty in hydrologic prediction. Water Resources Research, 43(1).

Ajay Kalra, Sajjad Ahmad. (2009). Using oceanic‐atmospheric oscillations for long lead time streamflow forecasting. Water Resources Research, 45(3).

Alp, M., Cigizoglu, H. K. (2007). Suspended sediment load simulation by two artificial neural network methods using hydrometeorological data. Environmental Modelling Software, 22(1), 2-13.

Applegate Jr, R. W., Squier, J., Vestad, T., Oakey, J., Marr, D. W., Bado, P., ... Said, A. A. (2006). Microfluidic sorting system based on optical waveguide integration and diode laser bar trapping. Lab on a Chip, 6(3), 422-426.