A generalized overview of artificial neural network and genetic algorithm

Abstract

Artificial Neural Network (ANN) and Genetic Algorithms are now widely used in various scientific problems including hydrology, and the results from the studies recommend more in depth studies in this regard. A neural network generally summed the weighted data of the input variables and with the help of some activation function, output variable is tried to be estimated. The advantages of such networks mainly lie in their ability to learn problems from the observed dataset of the variables. The network structure, number of layers and nodes, activation function varies with variation in problem domains which the model learns from the data with which it was trained. The drawbacks of such models are their disability to predict data for a relation which was not included in the training data which in turn make the model nonflexible. In addition, the required computation energy and model development time are little more than the conceptual hydrologic models. The selection of network structure, training algorithms, and activation functions are determined with the help of trial and error procedures. Variable dataset of different scales also reduces the model performance. The present note tries to give an overview of the different structures, activation functions, learning process of neural network, and the use of genetic algorithm to remove the trial and error methodology of selecting the ideal model structures and configurations Application of the iteration technique is also discussed. © 2010 Springer Science+Business Media B.V.