| Summary: | 碩士 === 國立中興大學 === 水土保持學系所 === 95 === The supply and demand of water resource is imbalance in Taiwan, if the reservoir management agency can predict accurately the inflow of reservoir from its up stream river, it should be helpfull for regulation and control of the reservoir.
This research collects the historical daily and 10-day inflow of reservoir watersheds, and uses time series analysis methods, its theory is already to be developed completely, to analyse its periodicity and secular trend separately, by way of autoregressive integrated moving average (ARIMA) models infer optimization parameter, for simulating and predicting the inflow of reservoir watersheds. On the other hand, the best way in which two series obtain assays and proves, compare the well and poor situated results with the observed inflow data, use to help the efficient operation of the reservoir water resources.
This research uses the inflow data of Wu-Seh (2) hydrologic station of Wu-seh reservoir of Jhuo-Shuei river basin science January of 1985 till December of 2005. The analyzed step was as following:
A. Make the daily and 10-day inflow series in order and file, then examine whether it is belong to white noise or not.
B. Carry on the daily and 10-day inflow series to be normalized and differential operated separately, then, draw the seasonality, circulation cycle and secular trend of inflow.
C. Use the daily and 10-day inflow series science 1985 till 2004, finally to set up the ARIMA(p, i, q)×(P, I, Q)k model.
D. Lastly, I tried to predict the inflow data of the 21-th year with both the erected two model and then compared it with the observed inflow data in 2005 year, by way of the best two ARIMA model.
Three conclusiones were found as following:
1. To adopt MA model for simulation and prediction of the daily inflow series is not so well as expectancy for this series’s randomness is so high that both of its autocorrelations and partial autocorrelations can not convergence.
2. Relatively, because the 10-day inflow series shows as characteristics of seasonality, periodicity and randomness, that means the factors are complicated, and will influence the series apparently, so to adopt the multiply, high rank seasonal ARIMA model for simulating and predicting the 10-day inflow series, it is found that the prediction ability of the best model is very well than the daily inflow series.
3. This study used 10-day as the time unit to predict the inflow series. Not only it can obey the norm curve but also can afford a concrete and practical analysis tool, ARIMA, and then can predict correctly an inflow from the upstream creek to the reservoir. It can be an authority for supporting the reservoir agency to adjust and control the water resources.
|
|---|
