Author | Nikhom Saterngram |
Call Number | AIT Thesis no.WM-03-04 |
Subject(s) | Neural networks (Computer science) Rain and rainfall--Sirikit reservoir--Forecasting
|
Note | A thesis submitted in partial fulfillment of the requirements for the
degree of the Master of Engineering |
Publisher | Asian Institute of Technology |
Abstract | Accurate forecast of inflows to reservoir is of particular interest for reservoir
operation and scheduling. For particular reservoir equipped with control gates,
improved criteria for gates operation during flood period can be assessed if reliable
forecasts of inflows to the reservoir are available. Prediction of rainfall should be
considered for the extension of lead-time of forecast. This study presents the application
of the Artificial Neural Networks (ANN) for daily rainfall time series forecasting in Nan
river basin and prediction of inflows to Sirikit reservoir and daily streamflow at main
gagging station along Nan river.
The Nan river basin is situated in the northern part of Thailand and drains water
from the upper basin to Sirikit reservoir. The characteristic of the upper part of Nan
river basin is mountainous area which the width is relative small compared to the length
of the basin. Due to the reason, rainfall from the catchment contributes to runoff rather
quickly and sometimes cause flood. At present, real time operation of Sirikit reservoir
in Nan river during flood periods frequently causes severe flooding in downstream areas
especially in the vicinity of Uttradit and Phitsanulok provinces. The excessive release
of Sirikit reservoir is due to lacking of information of local inflows from downstream
tributaries of Nan river. The success of flood forecasting for more than three days in
advance in the Nan river was found to depend heavily on the accuracy of rainfall
forecast. Hence, rainfall forecast is necessary for an extension of lead-time of food
forecast beyond 3 days in advance.
The ANN models (WinNN32) have been applied for daily rainfall time series
forecasting based on the Singular-Spectrum Analysis (SSA) with Principle Component
Analysis technique. The ANN models cannot be directly applied to forecast
discontinuous signals, like rainfall time series, as the universal function approximation
theorems for neural networks system that requires the continuity of the function to be
approximated. In order to avoid the effect of discontinuous of a signal, the SSA
technique is applied to forecast the signal by decomposing the raw rainfall time series
into reconstructed components. Then, reconstructed components are simulated by
ANNs to obtain forecasted rainfall time series for several lead-times (one day, two days
and three days in advance). After that, forecasted rainfall is obtained as inputs for
prediction of inflows to the Sirikit reservoir and streamflow at main gauging stations
along the Nan river downstream of the Sirikit dam.
The results of the forecast of inflows to Sirikit reservoir and streamflow at
downstream stations are found to be satisfactory for 1-5 days in advance. |
Year | 2004 |
Type | Thesis |
School | School of Engineering and Technology (SET) |
Department | Department of Civil and Infrastucture Engineering (DCIE) |
Academic Program/FoS | Water Engineering and Management (WM) |
Chairperson(s) | Tawatchai Tingsanchali |
Examination Committee(s) | Babel, Mukand Singh ;Dutta, Dushmanta |
Scholarship Donor(s) | The Oil Refinery Contract Contribution Fund
Committee, Energy Policy and Planning Office
Bangkok, Thailand |
Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2004 |