Investigation of artificial neural network models for streamflow forecasting

Full text for this resource is not available from the Research Repository.

Tran, Huu Dung, Muttil, Nitin and Perera, B. J. C (2011) Investigation of artificial neural network models for streamflow forecasting. In: MODSIM2011, 19th International Congress on Modelling and Simulation. Chan, F, Marinova, D and Anderssen, RS, eds. Modelling and Simulation Society of Australia and New Zealand, Australia, pp. 1099-1105.

Abstract

Time series forecasting is the use of a model to forecast future events based on known past events. Accurate forecasts of time series variables at different time scales are becoming increasingly necessary to facilitate mitigation of negative impacts of climate change, maximisation of system benefits from improved planning and management, and minimisation of system failure risks in all social, economical and environmental activities. Examples include hourly forecasts of rainfall (1-48 hours ahead), which are useful for flood warning, and monthly forecasts of streamflow (1-12 months ahead), which are beneficial for planning and operation of water supply systems. There are many other applications in finance and economics such as tourism demand forecasting and stock forecasting. Artificial neural network (ANN) models, which are considered as a category of the data-driven techniques, have been widely used in streamflow forecasting. Several distinguishing features of ANN models make them valuable and attractive for forecasting tasks. First, there are few a priori assumptions about the models as opposed to model-driven techniques. They learn from examples and capture the functional relationships among the data even if the underlying relationships are too complex to specify. Second, ANN models can generalize after learning from the sample data presented to them. Third, ANN models are universal functional approximators for any continuous function to the desired accuracy. Fourth, ANN models have flexible structures that allow multi-input and multi-output modelling. This is particularly important in streamflow forecasting where inflows at multiple locations are considered within a catchment. This paper investigated two basic ANN models, namely, feed forward neural network (FFNN) and layered recurrent neural network (LRNN) for streamflow forecasting in an attempt to understand why ANN models were used successfully in some streamflow forecasting studies but not always. In our study, two hypothetical and two real datasets were used to test performance of two different ANN models using feed forward and layered recurrent structures. Furthermore, an existing input selection technique using partial mutual information (PMI) approach, which can remove the insignificant inputs and thus potentially enhance the performance of ANN models, is also investigated. The results showed that the PMI approach correctly identified significant inputs of the two hypothetical datasets. However, the forecasting performance of FFNN and LRNN were not enhanced, when PMI identified inputs were used in comparison to using all inputs. The LRNN did not outperform the FFNN, although it is expected to perform better. Performance of both FFNN and LRNN models are related to noise level and autoregressive feature of time series data.

Additional Information

The 19th International Congress on Modelling and Simulation (MODSIM2011) was held at the Perth Convention and Exhibition Centre in Perth, Western Australia, from 12 to 16 December 2011

Item type Book Section
URI https://vuir.vu.edu.au/id/eprint/9530
Official URL http://www.mssanz.org.au/modsim2011/C1/tran.pdf
ISBN 9780987214317
Subjects Historical > Faculty/School/Research Centre/Department > School of Engineering and Science
Historical > FOR Classification > 0905 Civil Engineering
Historical > SEO Classification > 9609 Land and Water Management
Keywords ResPubID23193, artificial neural networks, streamflow forecasting, input selection
Citations in Scopus 5 - View on Scopus
Download/View statistics View download statistics for this item

Search Google Scholar

Repository staff login