以類神經模糊推理修正水位預測之一階延遲現象

Revising One Time Lag of Water Level Forecasting with Neural Fuzzy System

鐘侑達(You-Da Jhong) ； 黃晉華(Chin-Hua Huang) ； 陳昶憲(Chang-Shian Chen) ； 蘇惠珍(Hui-Chen Su)

《農業工程學報》 57卷3期 (2011/09) Pp. 1-16

https://doi.org/10.29974/JTAE.201109.0001

摘要

台灣每年的颱風暴雨之雨量強度高且總雨量十分驚人，往往造成下游洪水氾濫，損失難以估計，因此對於河川水位之預報已成爲重要研究課題。由於颱風暴雨河川水位變化與前一時刻之水位關係具有高度相關性，故時間序列分析方法在預報研究上經常被採用。基於數學模型的限制，時間序列在預測上常產生一階時刻延遲(one time lag)現象，因此將造成模式對於預測最高水位到達時刻產生誤差。故此本研究嘗試加入調適性網路模糊推論系統(Adaptive-Network Based Fuzzy Inference system; ANFIS)，針對時間序列在水文資料預測上進行修正。研究中ANFIS將利用雨量跟水位資料進行建模，由雨量序列資料提前推估水位變化趨勢，以修正時間序列一階時刻延遲問題；期望藉由此流程改善時間序列對於最高水位到達時刻所產生的誤差。研究亦同時建立一組ANFIS模式，探討利用雨量對水位進行推估可行性。由研究結果得知，ANFIS直接藉雨量模擬水位的準確件不佳，但應用在藉由雨量變化趨勢取得水位變化趨勢，以修正時間序列對最高水位到達時刻的預報，則可得到較準確與穩健的結果。

關鍵字

時間序列 ； 調適性網路模糊推論系統

並列摘要

Taiwan experiences quite high intensity of rainfalls in typhoons every year. It has often caused flooding in the downstream of rivers and huge loss in economics. Thus, the prediction of rainfall-runoff is a very important issue. Because the water level changes of storms significantly possess a genetic effect, the time series model has often been adopted foe forecasting (lie water level in the past. However, analyses using time series often result in one lime lag phenomenon and leading to errors in forecasting the arrival time of the peak water level due to the limitation of the mathematical model. Therefore, in this study, efforts have been made to improve tile time series model by involving the Adaptive-Network Based Fuzzy inference System (ANFIS) in the analyses of hydrological data. In the ANFIS, rainfall and water level data will be utilized to set up the model and to correct the lime lag problem by virtue of rainfall series data for predicting the trend of the water level change beforehand. It is hoped that the modified procedures proposed in this study will correct the errors caused by the time series in predicting the arrival time of the peak water level, in addition, the feasibility of estimating the water level through the rainfall will be investigated by virtue of the built ANFIS model. Results obtained from this study reveal that the ANFIS can not simulate the water level with good accuracy by using the rainfall, but it can yield better accuracy and stable results in comparison to the time series model in predicting the arrival time of the highest water level by virtue of the trend of the water level change obtained from the trend of the rainfall change.

並列關鍵字

Time Series ； Adaptive-Network Based Fuzzy Inference system