智慧型抽水站排水系統水位預報及操作策略整合模式

Abstract

高度開發城市因面臨都市化與氣候變遷導致洪峰流量快速上升之影響，使得都市防洪成為其重要議題。因此，為減輕未來洪水帶來損失，建立一有效率且準確之模式以預測洪汛時期市區抽水站之內外水位與指引適當抽水機操作有其迫切性與重要性。近年來，人工智慧技術運用於處理高度複雜之非線性系統有出色的處理能力，本研究運用該技術於即時水位預報等模式上。本研究選定臺北市玉成抽水站為研究對象，研究目標為建置即時水位預報和抽水機組操作策略兩種模式。 本研究首先蒐集區域內水文計量站之歷史紀錄，通過相關性及歷程分析搜尋降雨逕流之稽延關係，並運用Gamma Test由多種雨量資訊篩選影響水位之重要因子，進而決定模式輸入因子。模式建置部分，使用一種靜態類神經網路(倒傳遞類神經網路─BPNN)，二種動態類神經網路(隱藏層回饋式類神經網路─Elman NN, 非線性自回歸與外部輸入類神經網路─NARX)建立多時刻內水位預報模式；外水位預報模式則以三階自迴歸模型(AR3)建立，並比較評估水位預報模式之準確度與穩定性。本研究第二部分將著重於抽水站抽水機組操作策略，蒐集其操作之歷史紀錄，包含抽水機組、重力閘門操作紀錄，並配合水位預報模式之輸出水位值進行相關性及歷程分析，探討影響抽水機組操作之機制。最後，調適性網路模糊推論系統(ANFIS)負責整合水位預報模型建立多時刻抽水機組操作策略模式。 結果顯示，動態類神經網路NARX訓練之內水位預報模式表現優於另外兩種網路；AR3模式能於1小時內準確推估外水位；另ANFIS所整合之抽水機組操作模式能精確且穩定模擬歷史人為操作歷程。本研究所提出之模式與方案，在洪汛時期可提供各抽水站中操作人員與決策者適當且即時之抽水機組操作策略，進而減輕洪災以達成都市防洪之目的。

Urban flood control is a crucial task in developed cities, which faces a great challenge of fast rising peak flows resulting from urbanization and climate change. Therefore, it is imperative to construct an efficient and accurate model to forecast the water levels of rivers in urban areas during flood periods. Artificial intelligence (AI) techniques possess an outstanding ability to handle highly non-linear complex systems and are implemented to make real-time water level forecasts in this study. The Yu-Cheng pumping station located in Taipei City, Taiwan, is selected as the study area. The purpose of this study is to construct water-level forecasting models and a real-time operating strategy of pumps. The first part of the study is dedicated to water level forecasting. Hydrological data were collected and fully explored by statistical techniques to identify the time span of rainfall affecting the rise of the water level in the floodwater storage pond (FSP) at the pumping station. Effective factors (rainfall stations) that significantly affect the FSP water level are extracted by the Gamma test (GT). For model construction, one static artificial neural network (ANN) (backpropagation neural network-BPNN) and two dynamic ANNs (Elman neural network - Elman NN; nonlinear autoregressive network with exogenous inputs - NARX network) are used to construct multi-step-ahead FSP water level forecasting models. A third-order autoregressive model (AR3) is used to construct multi-step-ahead river water level forecasting models. The second part of this study is dedicated to operating strategy. Historical operating records of pumps at the pumping station were collected. Effective factors that affect the pumping operation are explored through the correlation analysis between the forecasted FSP and river water levels and historical hydrographs and operating records. The adaptive network-based fuzzy inference system (ANFIS) is used to construct multi-step-ahead operation strategy models in consideration of water level forecasting models. The results demonstrate that the dynamic NARX network is superior to the two other comparative ANN models in making FSP water level forecasting; the AR3 model can accurately make river water level forecasting; and the ANFIS can suggest an operating strategy that suitably and reliably simulates historical pumping operations. The proposed methodology can provide managers and operators of pumping stations with a guideline for making suitable real-time pumping operation in response to drastic water level variations during flood periods, which is beneficial to urban flood control management.