以類神經網路為基礎的X3D虛擬實境模擬水庫即時操作----以石門水庫為例

Simulation of Real-Time Operation by ANN-Based X3D Virtual Reality for the Shihmen Reservoir

李婉君(Wan-Jyun Li)

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/碩士(2008年)

https://doi.org/10.6342/NTU.2008.01714

摘要

為使水庫之營運在防汛時期有精確的操作策略與視覺化之操作介面，本研究建立了一種模擬操作模式與兩種即時操作模式供水庫操作參考；模擬操作乃根據前人以遺傳演算法(Genetic Algorithms, GA)所得到之最佳操作歷程，設計ㄧ洩放流程，模擬水庫各個設施實際可行之操作方式，亦檢測了不同閘門開度條件下所得之結果；研究顯示，總洩放量與期末水位幾乎可達到GA所設之目標，模擬的成效相當良好。在即時操作部份，本研究提供了兩種模式，第一種操作僅以目標水位為導向，不考慮複雜的演算法；第二種操作則是以GA之結果，做為倒傳遞類神經網路之訓練標的，推估下一時刻目標水位。即時操作的結果顯示，以目標水位為標的的操作模式較為保守，能夠使水位快速的達到所期望之高程，但在洪峰消減率上則表現較不佳；而倒傳遞類神經網路操作之結果不論在洪峰消減率或是期末水位的表現上，皆較原操作佳，能有效改善原操作的不確定性及不平滑性。此外，為使研究成果能有效展示，本文應用新一代網路三維標準格式X3D(Extensible 3D, X3D)作為展示介面。X3D的格式不僅可在水文資訊系統上達到快速的資料傳遞整合工作，更可讓決策者以視覺化的方式看到資料的即時變化情形及相關水庫操作決策之影響，是結合水文系統分析及網路資訊設計的ㄧ大趨勢，可做為水文資訊平台之應用。

關鍵字

類神經網路 ； 水庫防洪即時操作 ； X3D ； 虛擬實境

並列摘要

In order to develop a more precise strategy and a visional interface for the management of reservoir during typhoon or heavy rainfall events, this study provides a simulated control model and two real-time control models for reservoir operation. The flood release process has been designed in simulation control model to simulate the feasible operations of every reservoir facility according to the best result of Genetic Algorithms(GA) obtained from a previous study. Besides, performances from different level of open-gap in the release gate are verified. The results indicate that the simulated outcomes are fairly good and both total amount of outflow and the water demand are similar as those searched by GA. This study provides two real-time control models; the first one is merely derived from target water levels, and no complex algorithms; the second real-time control model uses the results simulated by GA as training patterns for Back Propagation Network(BPN)to estimate water release at next hour. Results from first model show the water level can rapidly reach the designed target but performance on the reduction of flood peak rate is not good enough. Whereas, results gained from BPN model not only produce better reduction of flood peak rate but also reach the target at the end of the flood. Moreover, both uncertainty and smoothly existed in original reservoir operation can be effectively improved through BPN model. In addition, a new generation of three-dimensional standard format X3D(Extensible 3D, X3D)is applied herein to display the on-line operating results. X3D can efficiently achieve the goal of data communication and integration in hydroinformatic system. Furthermore, the variation of data structure and the impact of operational strategy can be easily received and understood by decision maker in visional way. It is a tendency to combine hydrological analysis and information design on the Web.

並列關鍵字

ANNs ； Reservoir real-time flood Operation ； X3D ； Virtual Reality

參考文獻

蔡斐毓，2006，類神經網路於颱風降雨與流量預報之研究，國立台灣大學土木工程學研究所博士論文。

張郁麟，2007，倒傳遞類神經網路應用於台灣北部水庫懸浮固體濃度即時分析與預測之研究. 國立臺灣大學生物環境系統工程學系暨研究所碩士論文。

張斐章、張雅婷、張麗秋，2004，結合人工智慧與專家知識之智慧型水庫操作系統. 農業工程學報, 50(4)：14-27。

王國威，2006，運用懲罰機制遺傳演算法於水庫洪颱操作之規則，淡江大學水資源及環境工程學系研究所碩士論文。

張雅婷， 2006，調適性網路模糊推論系統於水庫操作之研究，國立臺灣大學生物環境系統工程學系暨研究所碩士論文。