HR: 17:45h
AN: H54B-08    [Abstracts]
TI: A New Integrated Neural Network Architecture for Streamflow Forecasting
AU: * Teegavarapu, R S
EM: ramesh@engr.uky.edu
AF: University of Kentucky, 161 Oliver Raymond Hall Department of Civil Engineering, Lexington, KY 40506 United States
AB: Streamflow time series often provide valuable insights into the underlying physical processes that govern response of any watershed. Patterns derived from time series based on repeated structures within these series can be beneficial for developing new or improved data-driven forecasting models. Data-driven models, artificial neural networks (ANN), are developed in the current study for streamflow prediction using input structures that are classified into geometrically similar patterns. The number of patterns that are identified in a series depends on the lagged values of streamflows used in the input structures of the ANN model. A new modular and integrated ANN architecture that combines several ANN models, referred to as pattern-classified neural network (PCNN), is proposed, developed and investigated in this study. The ANN models are used for one step-ahead prediction of streamflows for Reed Creek and Little River, Virginia. Results obtained from this study suggest that the use of these patterns in the process of training has improved the performance of the neural networks in prediction. The improved performance of the ANN models can be attributed to prior classification of data, which in a way has complimented and enhanced the already existing classification abilities of the neural networks. The PCNN architecture also provides the benefit of better generalization of a data-driven model by developing several independent models instead of one global data-driven prediction model for the entire data.
DE: 1800 HYDROLOGY
DE: 1821 Floods
DE: 1829 Groundwater hydrology
DE: 1857 Reservoirs (surface)
DE: 1860 Streamflow
SC: Hydrology [H]
MN: Fall Meeting 2005