Geology Reference
In-Depth Information
(auto-regressive moving average with exogenous inputs) models [ 16 ] have gained
more attention because of their satisfactory prediction performance and easy
implementation procedure. The research conducted by [ 17 , 88 ] has demonstrated
the success of these linear models in different applications. Inability to represent the
nonlinear dynamics inherent in hydrological processes was considered as the
serious disadvantage of the above mentioned models [ 44 ]. The researchers quest for
models which incorporate nonlinearity of the system with relatively short imple-
mentation effort, led hydrology to nonlinear pattern recognition and system control
theory borrowed from electronics and communication engineering stream. In the
early 1990s, much research was carried out in hydrology utilising the capabilities of
advanced nonlinear system theoretical modelling approach called Arti
cial Neural
Networks (ANN) [ 35 , 50 ].
The advent of arti
cial
intelligent
techniques in hydrology brought a new
dimension to
flood modelling [ 18 , 39 , 40 ]. Among several arti
cial intelligence
methods, arti
cial neural networks (ANN) holds a vital role and ASCE Task
Committee Reports [ 11 , 12 ] have accepted ANN as an ef
cient forecasting and
modelling tool. Over the last decade, the arti
cial neural network has gained great
attention and has evolved as the main branch of arti
cial intelligence that is now a
recognized tool for modelling the underlying complexities in many arti
cial or
physical systems including
oods [ 2 , 86 ]. Unlike traditional conceptual and physics
based models, Artificial Neural Networks are able to mimic flow observations,
without any mathematical descriptions of the relevant physical processes. A study
by Jain et al. [ 46 ] demonstrated that the distributed structure of the ANN was able
to capture certain physical properties. The success of hydrological forecasting
systems depends on accurate predictions in the longer forecast lead time. Multi-
step-ahead prediction is a challenging task which attempts to make predictions
several time steps into the future. Dawson and Wilby [ 23 ] focused into neural
network application on rainfall-runoff modelling and stream
flow modelling. Maier
et al. [ 61 ] provided a good review of neural network models used since 2000 for
water quantity and quality modelling. Chang et al. [ 21 ] developed a two-step-ahead
recurrent neural network for stream
flow forecasting. Later, they explored three
types of multi-step ahead (MSA) neural networks viz. multi-input multi-output
(MIMO), multi-input single-output (MISO) and serial-propagated structure for
rainfall-runoff modelling using data sets from two watersheds in Taiwan [ 22 ].
Nayak et al. [ 69 ] gave a detailed review of the application of ANFIS in rainfall
runoff modelling. Mukherjee et al. [ 66 ] points out the advantages of support vector
machines (SVMs) in making better predictions than other approximation methods
such as polynomial and rational approximation, local polynomial techniques and
arti
cial neural networks. A comprehensive review by Abrahart [ 1 ] provided two
decades of neural network rainfall-runoff and stream
ow modelling and suggested
extended opportunities in this
field.
 
Search WWH ::




Custom Search