Effect of data structure for time series prediction using time delay neural network

Neural networks with their adaptive and nonlinear properties have been used in different applications, in particular, for time series prediction is very common by several field researchers. In past decade years, many neural network models have been developed according to its topology structure and message transport ways, the stability, convergence and generalization ability of these models also have been proved by the scientist and experts on neural network theory. However, the engineers and practical user has necessary to know properties of forecasting problem, in other words, data pre-processing should be emphasized in order to obtain more available forecast results. This paper focuses on comparison of time series forecasting results using time delay neural network (TDNN) based on Genetic Algorithms (GA) according to different data structures of a set of field data in order to emphasize the importance of data pre-processing.