以遺傳規劃建構智慧型洪水預報機制

Establishing Intelligent Flood Forecasting Mechanism by Genetic Programming

陳昶憲(Chang-Shian Chen) ； 鍾侑達(You-Da Jhong) ； 方唯鈞(Wei-Chun Fang) ； 葉昭憲(Chao-Hsien Yeh)

《中國土木水利工程學刊》 23卷3期 (2011/09) Pp. 317-327

https://doi.org/10.6652/JoCICHE.201109_23(3).0008

摘要

遺傳規劃（Genetic programming, GP）可模擬複雜的非線性關係，然而單一定率 （deterministic）預測模式之推估，為適應所有模擬案例，將使其推論有趨向平均化的結果，較無法針對特殊或極端的事件去做個別探討，故其對於未知事件的預測未必能達到檢定之水準，也增加了颱洪預測的風險。為避免使用單一模式預測之風險，本研究嘗試根據颱風資料特性，以遺傳規劃模式建構知識庫╱模式庫，經由推論機制，依照輸入結構的不同選擇模式庫中之對應模式，增加時空變異的掌握能力。另亦考慮到不確定性，於輸出值運算結束後，利用自適應網路模糊推論系統（Adaptive Network based Fuzzy Inference System, ANFIS）推估模式修正可能發生之誤差並加以修正。本文建立之遺傳規劃模式庫應用於烏溪流域颱風水位之預報上，其驗證結果在預測整體表現具有穩定性及準確性。

關鍵字

遺傳規劃 ； 水位推估 ； 適應性類神經模糊推估系統

並列摘要

Genetic programming can simulate the system with complex nonlinear relationships among its components. However, the estimation by a single deterministic model trends to have the result of averaged inferring because the model intends to adapt all the training cases. In this way, the model not only fails to simulate the special or extreme events individually, it also increases the failure risk of the typhoon or storm forecasting to reach certain validation standards. To avoid the risk of this potential forecasting failure, this study suggested a genetic programming knowledge/model base which can select an appropriate model to improve the adaptability on the temporal-spatial variations by the inference mechanism based on the input structure from the typhoon information. To consider the uncertainty, the output from the model will be revised to reduce the possible error through an Adaptive Network based Fuzzy Inference System (ANFIS). Finally, this genetic programming model base was applied to the flood stage forecasting of Wu-Xi watershed for typhoon events, and the robust and accuracy of this model on the stage forecast were demonstrated by the verification results.

並列關鍵字

genetic programming ； flood forecasting ； adaptive network based fuzzy inference system