A Study on Reservoir Inflow Forecasting Using Genetic Programming
Resource
臺灣水利,58(1),1-9
Journal
臺灣水利
Journal Volume
58
Journal Issue
1
Pages
1-9
Date Issued
2010-03
Date
2010-03
Author(s)
Chen, Yung-Hsiang
Abstract
流量預測對於水資源系統規劃及操作甚為重要,尤以中、長期流量預測對於乾旱時期供水系統為甚。而在各種供水系統中,水庫向來被視為最重要且有效之蓄水設施,其入流量如能於事前精確地預測,將有助於水庫操作與管理。惟因流量之形成具高度非線性及時變性,且隨空間分布而異,故甚難預測。近來有甚多研究發展出不同之流量預測模式以求更精確之結果,而以達爾文進化論為理論基礎之遺傳規劃法(或稱基因規劃法)為其中較為嶄新之一種,由於該法具探索及學習隱藏於資料關係之能力,且能自動化地以數學運算元求解,並將模式以方程式展現,故已應用於多種領域。本研究之目的係將遺傳規劃法應用於週期性之流量預測,分別以石門水庫之旬入流量及集水區平均雨量資料進行不同之輸入組合,再藉由根均方差最小之目標函數,建構合適之入流量預測模式。研究結果顯示,遺傳規劃法所建構之預測模式表現較時間序列模式及線性迴歸模式為佳,且輸入資料經標準化後亦能增加模式之精確度。
Subjects
流量預測
旬流量
降雨-逕流模式
遺傳規劃法
石門水庫入流量
Streamflow forecasting
Ten-daily streamflow
Rainfall-runoff model
Genetic programming
Shihmen reservoir inflow
Description
Streamflow forecasting is of significant importance for planning and operation of water resource systems. Mid-term streamflow forecasting is especially important for the operation of water supply systems over drought seasons. Among the water supply systems, reservoirs should be regarded as the most important and effective water storage facilities. If the seasonal inflows of reservoir can be forecasted precisely beforehand, it may benefit the reservoir operation and management. However, streamflow formation is a highly non-linear, time varying, spatially distributed process and difficult to be forecasted. A large variety of models have been proposed to get more accurate and reliable performance for streamflow forecasting. Among the rainfall-runoff models, genetic programming (GP) has the advantages of its ability to learn relationships hidden in data and express them automatically in a mathematical manner. This study applied GP to establish the ten-daily inflow forecasting models for Shihmen Reservoir catchment. The forecasting of 10-day reservoir inflows reveals the excellent effectiveness of GP, and standardization is beneficial to modeling for seasonal time series.
Type
journal article
File(s)![Thumbnail Image]()
Loading...
Name
遺傳規劃法應用於水庫入流量預測之研究.pdf
Size
1.25 MB
Format
Adobe PDF
Checksum
(MD5):9e5594630d92586dd7330087761c47f2