| Summary: | 碩士 === 國立成功大學 === 水利及海洋工程學系 === 87 === Based on GM(1,1) of grey system, the back-propagation network(BPN) of artificial neural network and the conception of the threshold are used on hydrologic process forecasting.
Different thresholds are first used to turn the original daily rainfall data and the original 10 days streamflow data into new series. For daily rainfall, we can get four series for each threshold namely：the days of daily rainfall larger than the threshold; the days of daily rainfall smaller than the threshold; the number of days of daily rainfall larger than the threshold; the number of days daily rainfall smaller than the threshold. The same procedures are undertaken for streamflow as well.
To build GM(1,1) for forecasting those new series and BPN for simulating and forecasting those new series, the results of two different models are compared with the observation values.
It can be concluded that these predictions of GM(1,1) for rainfall and streamflow are satisfactory in the events of occuring days, not the time period of occurrence. Meanwhile, the method of BPN has less performance in the occuring days, but has better performance in the time period of occurrence.
|
|---|
