Record Details

Artificial neural networks for wave propagation

DSpace at IIT Bombay

View Archive Info
 
 
Field Value
 
Title Artificial neural networks for wave propagation
 
Creator LONDHE, SN
DEO, MC
 
Subject parabolic equation method
water-waves
prediction
wave propagation
wave attenuation
refraction-diffraction
neural networks
modular networks
 
Description Wave propagation around and inside a harbor is conventionally studied by numerically solving a representative equation of short-wave progression or by taking actual measurements on a physical model. Although the numerical schemes yield workable solutions, underlying assumptions as well as noticeable difference between the resulting estimations and actual measurements leave scope to employ alternative approaches. The current study is an attempt in that direction and is based on the approach of neural networks. Modeled to imitate the biological neural network prevalent in human brains, an artificial neural network represents interconnection of computational elements called neurons or nodes, each of which basically carries out the task of combining inputs, determining their strength by comparing the combination with a bias (or alternatively passing it through a non-linear function) and firing out the result in proportion to such a strength. The network is first trained with examples, the strengths of interconnections (or weights) are accordingly fixed and then it is readied for application to unseen inputs. The applications of neural networks have now spread across all disciplines of ocean engineering, namely, harbor, coastal, offshore and deep-ocean engineering, and are directed towards function approximation, optimization, system modeling including parameter predictions. Advantages of the ANN schemes are improved accuracy, ease in application, reduced data requirement and so on. In the present work a feed forward modular neural network was developed in order to estimate attenuation of wave heights along the approach channel of a harbor starting from seaward boundary and ending at the harbor entrance. The trained network was found to satisfactorily follow the expected trend of wave height attenuation along the harbor channel. When tested for unseen input it yielded values of wave heights close to the numerical and physical models. The network also properly simulated the effect of variation of wave period as well as that of angle of wave attack on wave attenuation.
 
Publisher COASTAL EDUCATION & RESEARCH FOUNDATION
 
Date 2011-07-20T10:33:09Z
2011-12-26T12:51:26Z
2011-12-27T05:36:18Z
2011-07-20T10:33:09Z
2011-12-26T12:51:26Z
2011-12-27T05:36:18Z
2004
 
Type Article
 
Identifier JOURNAL OF COASTAL RESEARCH, 20(4), 1061-1069
0749-0208
http://dx.doi.org/10.2112/03-0014.1
http://dspace.library.iitb.ac.in/xmlui/handle/10054/5442
http://hdl.handle.net/10054/5442
 
Language en