人工智慧應用於都市排水系統抽水站水位預測與最佳即時操作之研究

A study of applying artificial intelligence to forecasting water level of pumping station and optimization of real-time operation at city drainage system

黃建霖(Chien-Lin Huang)

指導教授 ： 徐年盛

國立臺灣大學/工學院/土木工程學研究所/碩士(2009年)

https://doi.org/10.6342/NTU.2009.03009

摘要

抽水站為都市排水防洪系統中最重要工具之一，當都市內水無法以重力排水或宣洩不及時，抽水站就負責將低窪地區之內水排除至外水之河川中，因此抽水站操作之效能與效率，實為肩負了防範都市淹水之重責大任。本研究之目的在建立一方法以供進行都市排水系統中抽水站前池水位之預報與抽水機組之最佳即時操作。研究中，首先應用倒傳遞類神經網路進行中港東抽水站前池水位之預報，其模式分別係以實測之雨量與水位資料、實測雨量與預測水位、預測雨量與預測水位等三種架構方式來組成，接著進行預測時距之敏感度分析，來決定水位預測模式之最佳架構方式與抽水站最佳即時操作之優選時距。研究第二部分為架構兩種抽水站抽水機組之即時操作模式，第一種操作模式為以歷史操作紀錄來建構即時操作之預測模式；第二種操作模式則首先架構抽水站最佳即時操作之優選模式，並以禁忌演算法配合前池水位預測模式，優選歷史颱風暴雨事件期間，抽水機組之最佳即時操作方式。本研究以中港大排為研究案例，由分析結果可知第二種模式相較歷史操作記錄有較低之前池水位與較高之抽水機開關效率，顯示以禁忌演算法優選最佳即時操作之模式，能同時達到防洪效能與經濟效益兼顧之目的。

關鍵字

人工智慧 ； 水位預測 ； 優選模式 ； 即時操作 ； 類神經網路 ； 禁忌演算法

並列摘要

Pumping stations are the most important tools for flood mitigation in an urban area. When rainwater could not be drained off by gravity immediately, the pumping station is responsible for draining water from low-lying area to the river. The efficiency of pumping station operation takes great responsibility for preventing the damage of flood in the city. According to the above-mentioned viewpoint, this paper focus on forecasting the water level at pumping station and the best real-time operation of pumping machine at the drainage system of the city. The purpose of this paper is to invent two pumping station operation models. First model uses pumping machines real-time operation forecasting model to obtain pumping machines operation method. Second model first we apply back propagation neural network to forecast the water level at the front pool of Chung-Kong East pumping station. By the sensitivity analysis of the different time interval, we determine the best construction of water level forecasting model and the optimization time step of the real-time operation of the pumping station. In order to construct the optimization model of the real-time operation of the pumping station, we use tabu search and water level forecasting model to optimize the best real-time operation method of pumping machines in the historical typhoon and storm period. This research uses Chung-Kong-Da-Pai basin as a case study. The results of second model show that the lower water level of front pool of the pumping station and the higher efficiency of switch of the pumping machine, compared with the history operation record. It reveals that using tabu search to optimize the real-time operation can achieve the purpose of preventing the damage of flood and economic benefits at the same time.

並列關鍵字

artificial intelligence ； water level forecasting ； optimization model ； real-time operation ； neural network ； tabu search

參考文獻

17. 李翁碩，2007，抽水站水位預測及系統操作之研究，國立台灣大學生物環境系統工程研究所碩士論文。

18. 張凱堯、張斐章，2007，反傳遞模糊類神經網路於抽水站操作之應用，農業工程學報，53(1)：82-91。

20. 段鏞、傅金城、鄧慰先，2008，適應性網路模糊推論系統於降雨-洪水位預報之研究，台灣水利，56(2)：50-70。

11. 林永堂，2004，結合OLS與SGA建構輻狀基底類神經網路於洪

13. 邱建堯，2005，結合GA與CG優選最佳倒傳遞類神經網路-以雨水下水道水位預測模式為例，國立台灣大學生物環境系統工程研究所碩士論文。