以類神經網路進行灌溉排水水量預報之研究

隨著氣候變遷因素影響，極端降雨的現象愈趨明顯，且農業水資源常需因應臨時移用，貯水池調蓄能力更顯重要。桃園地區利用貯水池進行區域灌溉為特有農業灌溉方式，可因應桃園大圳、河水堰供水不足時輔助使用，如何永續發展和有效管理以強化農業水資源使用效率是重要課題。本研究在桃園大圳10支線與12支線以流域水文模式建構幹線水位預測模型，分別預測未來1至3小時水位，並轉換成水量，以期輔助決策管理系統。本研究結合物聯網雲端運算和Microsoft Azure Machine Learning平台中處理非線性問題，表現良好之方法「類神經網路」，降雨與水位之相互關係，以建置幹線水位預測模型，利用貯水池有效調配管理水資源，達到水資源永續利用之目的。結果顯示，水位預測之均方根誤差與平均絕對誤差均小於0.22公尺、相關係數均大於0.90且效率系數均大於0.80，表示此模式可提供良好的水位預測結果。未來，可根據監測系統接收到的降雨和水位預報提前掌握可調配的水量，以增加貯水池的水資源。

關鍵字

桃園大圳；水資源管理；物聯網雲端運算；類神經網路； Microsoft Azure Machine Learning

並列摘要

The climate change leads to increase the frequency of the extreme rainfall. How to enhance effective water use by sustainable development and effective management is a crucial issue. Taoyuan main canal transportation channel (TMCTC) is important to the agriculture water supply in Taoyuan and HsinChu County. It is necessary to have accurate hourly forecasting streamflow for water use in irrigated agriculture. Therefore, unlike the traditional water resources management, a novel forecasting model based on neural network regression model (NNR) is proposed to forecast the water level for 1 to 3 h lead time. The forecasts of the water level are transformed into the streamflow to allocate water resources. The proposed model is constructed using NNR from Microsoft Azure Machine Learning (Azure ML). The Azure ML is a kind of cloud computing, and IoT techniques, and NNR is great flexibility in modeling nonlinear processes. The results show that root mean square errors, coefficient of correlation and coefficient of efficiency are less than 0.22 m, greater than 0.9, and greater than 0.8, respectively. On the basis of the aforementioned performance measures, the proposed model can produce accurate forecasts, and is expected to be useful to water resources management.

並列關鍵字

Taoyuan main canal transportation channel ； water resources management ； cloud computing and IoT techniques ； neural network regression ； Microsoft Azure Machine Learning

國際替代計量

以類神經網路進行灌溉排水水量預報之研究

全文下載

主題瀏覽