Stochastic Simulation And Prediction Of Annual Runoff Series

The stochastic simulation of theory and methods of hydrological series are summarized in this thesis. Taking annual runoff of 11 stations in Shanbei area for example, firstly, the composition of the annual runoff series is analyzed, including identification and extraction deterministic cycle composition and the deterministic non-cycle composition. Secondly, the annual runoff series is simulated by linear stationary stochastic model and Perturbing Nearest Neighbor Bootstrap (PNNB) model. The most optimal stochastic models for annual runoff of 11 stations in Shanbei area are selected through the practical analysis of statistical parameters. Finally, annual runoff of 11 stations from 2001 to 2003 are forecasted by the optimal model. Main conclusions of the thesis are given as following.(1) Annual runoff of Shenmu station significantly shows a decreasing trend. The trend equation is y= -0.215t+441.33,1956≤t≤2003. Annual runoff of Zaoyuan station shows jumping variation. The jump is significant through a detailed diagnostic of the jumping identification. Due to significantance of 1996 year is the largest, the annual runoff series had occurred jumping variation in 1996 year. The mean of pre-jumped the annual runoff series is 0.79m~3/s. After the jumping, the mean of annual runoff series is 0.60m~3/s. the difference is 0.19m~3/s.(2) Annual runoff series of 11 stations in studied area have diversification in hydrological cycle. They all have small-scale hydrological cycle of 2 to 3years. Annual runoff series of Suide station, Xinghe station, Zhangcunyi station, Ansai station and Huangling station also have the hydrological cycle of 11 years. Annual runoff series of Liujiahe station, Jiaokouhe station, Zaoyuan station and Zhangcunyi station have the hydrological cycle of 22 years. In addition, for the length of the annual runoff series which is greater than 35 years, Maximum entropy spectrum analysis and spectrum analysis have similar resulst. But the length of the annual runoff series which is less than 35 years, power spectral analysis demonstrats the lack of low resolution. Thus, the maximum entropy spectrum analysis does have high resolution characteristics.(3) Optimal stochastic model of annual runoff series of 11 stations in studied area have been selected. AR (4) model can be used to fit annual runoff series of Suide station. Annual runoff series of Xinghe station is AR (2) model. AR (1) model is used to fit annual runoff series of Wuqi station, Shenmu station, Zhangcunyi station, Zhidan station, Ansai station, Huangling station, Liujiahe station, Jiaokouhe station and Zaoyuan station.(4) Pure random series is simulated by maximum entropy distribution model. If the annual runoff series follows approximate normal distribution. That is annual runoff series of the skewness which is small. Pure random series is simulated by maximum entropy distribution model and superposed to smooth model. Through the practical test, it can maintain the skewness of series. For the annual runoff series of relatively large skewness, pure random series is simulated by maximum entropy distribution model and is superposed to a smooth model to get annual runoff series. The skewness of series is generally smaller by the practical test.(5) Annual runoff series of 11 stations from 2001 to 2003 are forecasted. In 2002, annual runoff series of 11 stations except for Zaoyuan stations and Zhang cunyi station, predicted value of each station can meet the required long-term hydrological forecast accuracy. From 2001 to 2003, there are 15 predicted values of annual runoff series of 11 stations which can not meet the required long-term hydrological forecast accuracy. Among of 15 predicted values, 10 predicted values are greater than the measured values. It shows that the model is needed to be improved.(6) Annual runoff series is simulated by PNNB model. Through the practical test, this model can maintain the series of mean and variance. But it can not maintain the autocorrelation coefficient and skewness of the series well. It shows that the model is needed to be improved further.