人工智慧於都市防洪排水系統控制之研究

Artificial Intelligence for City Flood Control System

張凱堯(Kai-Yao Chang)

指導教授 ： 張斐章

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/博士(2009年)

https://doi.org/10.6342/NTU.2009.00988

摘要

雨水下水道系統為都市安全輸送暴雨逕流，減輕都市淹水風險的重要設施，然而都市化造成的高度開發，使降雨集水時間縮短，急遽暴雨在短時間內產生大量地表逕流無法及時排除，導致局部地區發生積水情形；而市區排水系統配合都市開發而逐漸地下化，使下水道系統水理狀況不易掌握，更增加都市防洪排水管理的困難度。本研究目的在透過人工智慧相關技術，模擬人類學習、適應、回想等能力，以解決都市排水複雜的非線性與時變性系統問題，建構市區雨水下水道水位預測模式及防洪抽水站即時操作指引模式，提供管理者進行都市防洪排水整體操控策略規劃之參考。本研究首先探討倒傳遞類神經網路(BPNN)於雨水下水道水位預測模式之應用，經測試不同時距輸入資料格式之多階段水位預測模式，成功建立預測準確性極高之雨水下水道水位預測模式，並探討得知短時距輸入資料格式建立之水位預測模式比長時距資料格式更能掌握未來時刻雨水下水道水位變化，此水位預測模式除為都市積水預警提供更寬裕的應變時間外，亦提供防洪抽水站操作對掌握未來時刻水位變化之需求。其次探討反傳遞模糊類神經網路(CFNN)及調適性模糊類神經網路(ANFIS)於都市防洪抽水站即時操作指引模式之應用，研究結果顯示，透過降雨量、前池水位變化、閘門啟閉及抽水機組操作等輸入資料，CFNN及ANIFS即能建置準確性極高的抽水操作指引模式；而ANFIS因具備高度學習能力，利用少數規則即可預測較CFNN更精確的未來時刻抽水機組操作需求，因此應用ANFIS建構之防洪抽水站即時操作指引模式，將可提供抽水站管理人員即時操作建議，提昇抽水站操作效率及安全性。

關鍵字

水位預測 ； 抽水站操作 ； 倒傳遞類神經網路(BPNN) ； 反傳遞類神經網路(CFNN) ； 調適性模糊類神經網路(ANFIS)

並列摘要

Drainage systems play an important role in transporting storm runoff and reducing flood risk in urban areas. Stormy water, discharged by underground drainage systems, is hard to control, especially in highly urbanized areas, where concentration time is shorten and runoff coefficients are increased. This study aims to construct water-level prediction models in urban drainage systems and real-time operational guidelines for flood control pumping stations by using artificial intelligent techniques (AI). The AI techniques could effectively solve highly non-linear control problems and robustly tune the complicated conversion of human intelligence to logical operating system.
This study first applies back-propagation neural networks (BPNN) to predict water-level in the urban drainage systems of Taipei city. The results show that BPNN could satisfyingly predict the water level with high accuracy. The model provides much longer responding time for urban flood management. The application also indicates that input data with shorter time interval has higher accuracy, which meets the need of pumping operation.
The real-time operation guidelines for pumping stations in urban areas are future investigated by using counterpropagation fuzzy-neural network (CFNN) and adaptive network-based fuzzy inference system (ANFIS). The results demonstrate that CFNN and ANFIS are both capable of forming reliable guidelines by using the information of precipitations, fore-bay water levels, gate operation and number of pumping station. It also indicates that ANFIS, comparing to CFNN, has better learning algorithm, which requires less rules to meet accuracy pumping operation needs. The real-time operation guidelines formed by ANFIS are recommended to managers for promoting operation efficiency and reliability.

並列關鍵字

Water-level prediction ； Pumping operation ； Back-propagation neural networks(BPNN) ； Counterpropatagation fuzzy neural network(CFNN) ； Adaptive network-based fuzzy inference system(ANFIS)

參考文獻

91.王安培、鄭博文，1998，“模糊決策在水庫操作之應用”，中原學報，26(3):17-25。

98.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。

99.林旭信，2004，「都市雨水下水道系統最佳化操作模擬」，國立台灣大學土木工程研究所博士論文。

105.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。

126.陳昶憲、楊朝仲、王益文，1996，“類神經網路於烏溪流域洪流預報之應用”，中華水土保持學報，27(4):267-274。