Devising a Two-Step Real-Time Recurrent Learning Algorithm for Streamflow Forecasting
Resource
台灣水利 50 (2): 15-21
Journal
台灣水利
Journal Issue
2
Pages
15-21
Date Issued
2002-06
Date
2002-06
Author(s)
Abstract
The architecture of Recurrent Neural Network (RNN) provides an efficient representation of dynamic internal feedback loops in the system to store information for later use. The Real-Time Recurrent Learning (RTRL) algorithm deriving for RNN is an on-line learning. The main feature of the RTRL is that it doesn't need a great deal of historical data for training, and it can perform on-line learning with training and operating at the same time. Most of the traditional physical models or feed-forward neural networks for streamflow forecasting emphasize the one-step ahead for ecasting. Due to the high mountain and steep channel all over the Taiwan island, the peak flow usually approach downstream about 2 or 3 hours after heavy rainfall on the watershed. Consequently, it is desired and will be very beneficial if the model can provide multi-step ahead forecasting. The main aim of this study is to develop a two-step ahead algorithm based RTRL. For the purpose of comparison, the predictive ability of RTRL and ARMAX model are performed by using the rainfall-runoff data of the Da-Chia River. Our results demonstrate that the RTRL has a learning capacity with high efficiency and is an adequate model for time series prediction.
Subjects
Recurrent neural network
Real-time recurrent learning algorithm
On-line learning
SDGs
Type
journal article
File(s)![Thumbnail Image]()
Loading...
Name
回饋式類神經網路於㆓階段即時流量預測.pdf
Size
539.38 KB
Format
Adobe PDF
Checksum
(MD5):fd8e42f634bd52ad111b34b7e91a1f77