Behnam Ababaei et al./ Elixir Comp. Sci. & Engg. 42 (2012) 6074-6077 6074
Introduction
Information about the parameters defining water resources
availability is a key factor in their management which improves
the operation policies for water resources systems. One of the
most important parameters in this area is river streamflow.
Moreover, climate change impact assessment studies often need
models capable in simulating river streamflow on a daily time
basis. The modeling of streamflow time process has essentially
followed two approaches (Razavi and Araghinejad 2009):
empirical and statistical simulation of the hydrological system.
In the first approach, the hydrological system is described by
theoretical and/or empirical (physical) relationships (e.g. Garrote
and Bras 1995), whereas, in the statistical approach, the
objective is to develop a model in order to represent the most
relevant statistical characteristics of the historical series
(Araghinejad et al. 2006).
Artificial Neural Networks (ANNs)
Many statistical-based methods are used to model and
forecast streamflow time series. In the recent years, Artificial
Neural Networks (ANNs), have been widely studied and applied
to simulate and forecast the hydrological variables (Hsu et al.
1995; Coulibaly et al. 2001; Razavi and Karamouz 2007;
Altunkaynak 2007; Razavi and Araghinejad 2009; Ahmed and
Sarma 2007; El-Shafiel et al. 2007).
Radial Basis Neural Network (RBN)
The radial basis function approach traces its roots from the
work of Powell (1987), whose use as an alternative tool to
learning in neural networks is particularly suited to multivariable
interpolation given irregularly positioned data points. Their use
in neural networks has found applications in solution of
classification problems, function approximation, noisy
interpolation, and regularization (Ke´gl et al. 2000) in various
engineering fields due to their advantages over traditional
multilayer perceptrons (Kagoda et al. 2010), namely faster
convergence, smaller extrapolation errors, and higher reliability
(Girosi and Pogio 1990). In hydrology and considering the
complex nature of the rainfall–runoff process which is usually
highly non-linear, the most suitable neural networks for
modeling the process should have the ability to approximate any
continuous function. The RBF technique provides good
generalization ability with a minimum number of nodes to avoid
unnecessarily lengthy calculations, in comparison with
multilayer perceptron networks (Moradkhani et al. 2004). The
architecture of radial basis function neural networks consists of
an input layer, one hidden layer and one output layer. Each node
in the hidden layer evaluates a radial basis function on the
incoming input. In this study, the radial basis function applied
was the Gaussian function and the neural network output was
then evaluated as the weighted linear summation of the radial
basis functions.
General Regression Neural Network (GRNN)
General Regression Neural Networks (GRNNs), falls into
the category of probabilistic neural networks. This neural
network like other probabilistic neural networks needs only a
fraction of the training samples a backpropagation neural
network would need. The use of a probabilistic neural network is
especially advantageous due to its ability to converge to the
underlying function of the data with only few training samples
available. The additional knowledge needed to get the fit in a
satisfying way is relatively small and can be done without
additional input by the user. This makes GRNN a very useful
tool to perform predictions and comparisons of system
performance in practice. GRNN consists of four layers namely,
input layer, pattern layer, summation layer and output layer
(Singh and Deo 2007). The first layer is fully connected to the
second pattern layer, where each unit represents a training
pattern and its output is a measure of the distance of the input
from the stored patterns. Each pattern layer unit is connected to
the two neurons in the summation layer: S-summation neuron
and D-summation neuron. The S-summation neuron computes
the sum of the weighted outputs of the pattern layer while the D-
summation neuron calculates the unweighted outputs of the
pattern neurons.
Elixir Comp. Sci. & Engg. 42 (2012) 6074-6077
Assessment of radial basis and generalized regression neural networks in daily
reservoir inflow simulation
Behnam Ababaei
*
, Teymour Sohrabi and Farhad Mirzaei
Department of Irrigation and Reclamation Engineering, University of Tehran, Iran.
ABSTRACT
In this study, two different type of Artificial Neural Networks (ANNs) were analyzed in
simulating the daily inflow into Taleghan reservoir in Iran. These types include: General
Regression Neural Network with standardized inputs (GRNN1) and with non-standardized
inputs (GRNN1), and Radial Basis Networks with standardized inputs (RBN1) and with
non-standardized inputs (RBN2). An iterative algorithm was designed to assess different
architecture of these models. Results revealed the potential of these models, as suitable tools
for simulating the daily reservoir inflow. Also, it was concluded that multiday averaging can
improve the simulation results considerably.
© 2012 Elixir All rights reserved.
Computer Science and Engineering
ARTICLE INFO
Article history:
Received: 2 November 2011;
Received in revised form:
15 December 2011;
Accepted: 26 December 2011;
Keywords
Taleghan Reservoir,
Daily Inflow,
Radial Basis Neural Networks,
Generalized Regression Neural
Networks.
Available online at www.elixirjournal.org
Tele:
E-mail addresses: Behnam.ab@gmail.com
© 2012 Elixir All rights reserved