Author(s): S. N. Londhe; Shalaka Shah
Linked Author(s):
Keywords: Rtificial neural networks; Knowledge extraction; Evaporation
Abstract: Artificial Neural Networks have been used increasingly in recent years for numerous hydrological applications because of their ability to represent non-linear relationships that are difficult to model by means of other computational methods (Sudheer and Jain, 2004). The major drawback of ANNs is that little is known about what is happening inside the ANN system. Thus ANNs are frequently criticized for operating as “black box” models where all dependencies (between parameters and responses) are hidden within neural network structure. As a result researchers have been trying to extract the knowledge from the trained ANNs, which will help them to become more widely accepted and reach their full potential as hydrological models. There are a few studies related to knowledge extraction in the field of hydrology which are pertaining to stream flow modelling (Sudheer 2005, Kalteh 2008, Jain and Kumar 2009). The proposed work will focus on preparing an evaporation model for a station, namely Nashik, using ANN and then extracting the knowledge locked in the weights and biases of the trained ANN. This will throw a light on working of ANN and its understanding of physics of the underlying process exploring the possibility of classifying the ANNs as “Grey” model rather than a complete “Black Box”. The evaporation will also be estimated using available empirical formulae which involve certain parameters and results will be compared. Additionally the results of knowledge extraction and parameters involved in these empirical models will also be discussed.
Year: 2016