Hydrologic Time Series Forecasting With Specialized Autoregressive Neural Networks

Author

• Paul Edward Roundy, Utah State University

Date of Award:

5-1999

Document Type:

Thesis

Degree Name:

Departmental Honors

Department:

Physics

Committee

Not Specified

Abstract

Interannual forecasts of hydrologic time series are important for water supply management. Specialized neural network algorithms for such forecasts are developed here. These methods are based on the principles of nonlinear dynamics and an autoregressive backpropogation training algorithm. The applicable nonlinear methods are described, as well as the algorithm. Some of the problems of these networks are summarized, along with some of their positive attributes. Tests showed that they may be able to learn and forecast some hydrologic data. These networks were most successful for forecasting the flows of the Colorado River. Some example forecasts are given, including the peaks of 1998 and 1999, as forecast in October, 1997.

Recommended Citation

Roundy, Paul Edward, "Hydrologic Time Series Forecasting With Specialized Autoregressive Neural Networks" (1999). All Graduate Theses and Dissertations, Spring 1920 to Summer 2023. 9009.
https://digitalcommons.usu.edu/etd/9009