遺傳演算法與類神經網路於地表地下聯合營運之應用

Applying Genetic Algorithm and Neural Network on the Conjunctive Use of Surface and Subsurface Water

張良正(Liang-Cheng Chang) ； 陳宇文(Yu-Wen Chen) ； 朱宏杰(Hone-Jay Chu) ； 黃浚瑋(Chun-Wei Huang)

《農業工程學報》 54卷2期 (2008/06) Pp. 81-93

https://doi.org/10.29974/JTAE.200806.0008

摘要

於傳統地表地下聯合營運管理模式中，常見的地下水反應方提式之處理方式可分爲崁入法(Embedding Method)與響應矩陳法(Response Matrix Method)崁入法是將一個地計水數值模式完繁嵌入，其計算量容易隨間題的增大而遽增；響應矩陣法則以線性矩陣方程式近似系統之反應行爲，然因其線性假設，並不適用於非線性之非拘限含水層管理問題。因此，本研究應用類神經網路學習地下水位變化，藕此建立一個可長期預測地下水位變化之類神經網路，並將其嵌入一個逐時刻優選之地表地下聯合營運模式，以遺傳演算法出連地表水與地下水兩系統，兩者間之調配原則採用「指標平衡」概念，無在維持地表地下互動的精神下，兩系統乃各自計算，地表水調配部分則以線性規劃求解，地下水系統之描述之類神經綱路。本研究進一步探討不同的營運方式對系統供水效能之影響，研究結果顯示，相較於地表水；水源易受水文條件之影響，地下水爲相對隱定的供水來源，在地下水系統旬平均洪水供水量僅佔總需量之7.48%（112.15萬噸／旬）案例中，聯合營運可以大幅降低缺水指數(SI)，最高可由2.93降低至0.79。此外，本模式透過不同的地下水分層設計，能夠實施不同強度之地下水使用策略，相較於「地表水先用地下水後供」之營運方式，可提供更彈性的地下水營運方式臁大幅降低整體缺水情形。

關鍵字

遺傳演算法 ； 類神經綱路 ； 指標平衡 ； 地表地下聯合營運

並列摘要

In the two traditional conjunctive use approaches, the embedded method and the response matrix method, are used to present groundwater system response. The embedded method embeds a groundwater numerical model in the management model but this method is usually associated with huge computational loading. The response matrix method approximates the system response as a linear matrix function but has no ability to deal with nonlinear problems in unconfined aquifer management. This study uses a BackPropagation Neural Network (BPN) which is a hypothetical way to simulate the nonlinear system response to groundwater management.
This study proposed a stepwise optimal management model for Conjunctive use of surface and subsurface water. The supply model for the conjunctive use problem was determined by a hybrid approach including Genetic Algorithm (GA), BPN and linear program (LP). GA was used to integrate a surface water system, constructed using LP, with a groundwater system, constructed using BPN. The operation rule curve and principle of 'index balance' were applied to supply water from both the surface water and subsurface water. The proposed model investigated the impact on operation strategies A small amount groundwater supply can significantly increase the water supply reliability.
Simulation results show that with only the groundwater supply increasing by 7.48% can significantly decrease the shortage index (SI). Besides, this work also demonstrates the 'rule curve' principle to control the priority of groundwater usage and is a flexible and effective strategy for the conjunctive use of surface and subsurface water.

並列關鍵字

Genetic algorithm ； Artificial neural network ； Principle of index balance ； Conjunctive use