Artificial Intelligent Systems and Machine Learning

G. S. Dwarakish ¹, Rakshith Shetty ¹, Usha Natesan ²

Affiliations
1 Department of Applied Mechanics and Hydraulics, National Institute of Technology Karnataka, Surathkal, P. O. Srinivasnagar, Mangalore 575 025, IN
2 Centre for Water Resources, Anna University, Chennai-600 025, IN

Abstract

Artificial Neural Networks (ANN) finds wide variety of application in solving problems related to coastal engineering. Its ability to learn highly complex interrelationship based on provided data sets with the help of a learning algorithm along with built in error tolerance and less amount of data requirement, makes it a powerful modeling tool in the research community. Large number of studies has been carried out in various fields like prediction of wave parameters, tidal level and storm surge, estimation of design parameters, liquefaction depth and scour depth to name a few. Various forecasting, estimation and supplement to the missing data studies carried out from different perspective ranging from, the sensitivity analysis to check the effect of input parameters and reduce the input size by discarding less effective ones; reducing the input size by using data assimilation techniques like principal component analysis to decrease the computational time requirement; usage of updated algorithms to overcome the problem of overfitting and overlearning, thereby increasing the network efficiency; has been carried out successfully, establishing ANN as an strong alternative to the data demanding and time consuming hydrodynamic and numerical models. As the validity of ANN to the ocean engineering applications became increasingly evident studies were incorporated in practical applications as well. Studies are being carried out to merge ANN with other AI techniques of Genetic Programming and Fuzzy Logic approaches to overcome the setbacks observed in ANN models. The studies have successfully shown that ANN can be applied to solve vast problems related to ocean engineering problems by meticulous selection of data, input parameters, network architecture and learning algorithms.

Keywords

Artificial Neural Networks, Artificial Intelligence, Coastal Engineering, Ocean Engineering.

Full Text

     

Gumageri Nagaraj ¹, G. S. Dwarakish ², Sreenivasulu Dandagala ², Usha Natesan ³

Affiliations
1 Department of Applied Mechanics and Hydraulics, National Institute of Technology, Karnataka, Surathkal, 575025, IN
2 Department of Applied Mechanics and Hydraulics, National Institute of Technology Karnataka, Surathkal, 575025, IN
3 Center for Water Resources, Anna University, Chennai-600025, IN

Abstract

Forecasting of wave parameters is necessary for many
marine and coastal operational related activities. In this paper, artificial neural network (ANN) as a robust data learning method is used to forecast the waveheight for the next 3hr, 6hr, 9hr, 12hr, 24hr, 48hr, 72hr, 96hr and 120hr in the Mangalore region, southwest coast of India. For this purpose two different models namely, Feed Forward Back Propagation (FFBP) and Nonlinear Auto Regressive Model with eXogenous input (NARX) of the ANN were used. The performances of developed models were evaluated using performance indices such as RMSE and CE. The CE values in FFBP model ranged from 0.997 to 0.785 while in NRAX model CE values are between 0.995 and 0.806 for the prediction time from 3hr to 120hr. A better agreement was observed between the observed and predicted waves for NRAX than that of FFBP for smaller (3-12hr) and larger lead period (24-120hr). Thus the NARX model performs better than the FFBP in terms of prediction capability and accuracy. 

Keywords

Waveheight, Prediction, ANN, FFBP, NRAX, RMSE.

Full Text