Computing:Theories and Applications (BICTA 2012),Advances in Intelligent Systems
and Computing 202,DOI:10.1007/9788132210412_1,Ó Springer India 2013
1
WaveletANN Model for River Sedimentation
Predictions
Raj Mohan Singh
1
1
Associate Professor, Department of Civil Engineering, MNNIT Allahabad, India
Email:
rajm@mnnit.ac.in
;
rajm.mnnit@gmail.com
Abstract. The observation of peak flows into river or stream system is not straight
forward but complex function of hydrology and geology. Accurate suspended se
diment prediction in rivers is an integral component of sustainable water resources
and environmental systems modeling. Agricultural fields’ fertility decays, rivers
capacity decreases and reservoirs are filled due to sedimentation. The observation
of suspended sediment flows into river or stream system is not straight forward
but complex function of hydrology and geology of the region. There are statistical
approaches to predict the suspended sediments in rivers. Development of models
based on temporal observations may improve understanding the underlying hydro
logical processes complex phenomena of river sedimentation. Present work uti
lized temporal patterns extracted from temporal observations of annual peak series
using wavelet theory. These patterns are then utilized by an artificial neural net
work (ANN). The waveletANN conjunction model is then able to predict the dai
ly sediment load. The application of the proposed methodology is illustrated with
real data.
Keywords: Wavelet analysis, ANN, WaveletANN, Time series modeling, Sus
pended sediment event prediction.
1 Introduction
Sediment amount which is carried by a river is complex functions of river’s
flow rate, and the characteristics of the catchment. Though discharge (flow rate)
can be measured at a site, determination of catchment characteristics accurately is
not always an easy task. The correlation between rivers’ flow and sediment obser
vation results and the basin’s characteristics should be determined for well plan
ning studies on soil and water resources development [1]. In rivers, a major part of
the sediment is transported in suspension. Recently, the importance of correct se
diment prediction, especially in floodprone areas, has increased significantly in
water resources and environmental engineering. A great deal of research has been
devoted to the simulation and prediction of river sediment yield and its dynamics
J.C.Bansal et al.(eds.),Proceedings of Seventh International Conference on BioInspired
[2][3]. The daily suspended sediment load (S) process is among one of the most
complex nonlinear hydrological and environmental phenomena to comprehend,
because it usually involves a number of interconnected elements [4]. Classical
models based on statistical approach such as multilinear regression (MLR) and se
diment rating curve (SRC) are widely used for suspended sediment modeling [5].
Hydrologic time series are generally autocorrelated. Autocorrelation in time se
ries such as streamflow usually arises from the effects of surface, soil, and
groundwater storages Sometimes significant autocorrelation may be the result of
trends and/or shifts in the series [6]. The application of ANN to suspended sedi
ment estimation and prediction has been recently used [7][8]. An ANN model
was also employed to estimate suspended sediment concentration (SSC) in rivers,
achieved by training the ANN model to extrapolate stream data collected from re
liable sources [9]. Bhattacharya et al. (2005) devised an algorithm for developing
a datadriven method to forecast total sediment transport rates using ANN [10].
Raghuwanshi et al. (2006) proposed an ANN model for runoff and sediment yield
modeling in the Nagwan watershed in India [11]. The ANN models performed
better than the linear regression models in predicting both runoff and sediment
yield on daily and weekly simulation scales.
Present work utilized temporal patterns extracted from temporal observations of
daily discharge and suspended sediment observed series using wavelet theory.
These patterns are then utilized by an artificial neural network (ANN). The wave
letANN conjunction model is then utilized to predict the yearly sediment flows in
a stream. The application of the proposed methodology is illustrated with real
data.
2 Artificial Neural Network
ANN is a broad term covering a large variety of network architecture, the most
common of which is a multilayer perceptron feedforwrd network (Fig. 1) with
backpropogation algorithm [12]. There is no definite formula that can be used to
calculate the number hidden layer(s) and number of nodes in the hidden layer(s)
before the training starts, and usually determined by trialanderror experimenta
tion.
The back propagation algorithm is used for training of the feed forward multi
layer perceptron using gradient descent, applied to sumofsquares error function.
This algorithm involves an iterative procedure for minimization of error function,
with adjustments to weights being in series of sequence of steps. There are two
distinct stages at each step. In the first stage errors are propagated backwards in
order to evaluate the derivatives of the error function with respect to weights. In
the second stage, the derivatives are used to compute the adjustments to be made
with weights [13] [14]. (Bishop, 1995; Singh et al., 2004). Present paper utilized
Levenberg Marquardt (LM) algorithm to optimize the weights and biases in the
network. LM algorithm is more powerful and faster than the conventional gradient
2
R. M. Singh
descent technique [15] [16]. Basics and details of ANN are available in literature
[17].
Enter the password to open this PDF file:
File name:

File size:

Title:

Author:

Subject:

Keywords:

Creation Date:

Modification Date:

Creator:

PDF Producer:

PDF Version:

Page Count:

Preparing document for printing…
0%
Commentaires 0
Connectezvous pour poster un commentaire