Nonparametric Disaggregation Models And Their Applications

Hydrology stochastic simulation is an important part of stochastic hydrology. A rational hydrological model is a kind of means and methods which has the great significance for the hydraulic engineering's planning, operation and management.This thesis quoted the methods which had given by Wang Wensheng and other scholars, taking the monthly runoff series and flood runoff series of 7 stations in Shaanbei region as examples, and applied nonparametric disaggregation models and improved nonparametric disaggregation model to their application of runoff simulation. The major results of the thesis are as follows.(1) Reviewed domestic and international progresses in hydrological stochastic simulation, and summarized the existing problems.(2) Reviewed the theory of kernel density estimation and univariate kernel density estimation models. Kernel function and bandwidth coefficient were selected and model orders were determined. Except first-order and second-order autocorrelation coefficient and Cs, this model can maintain good statistical properties of the observed series and characterize the process of annual runoff. The result is rationable.(3) Nonparametric disaggregation models. The practice of simulation monthly runoff using nonparametric disaggregation model shows that the model maintains good statistical properties of the observed series on mean, square deviation, Cv, Cs, maximum, minimum, first-order autocorrelation coefficient and second-order one, which in line with the hydrologic characteristics of watershed. Due to smooth of the kernel function, Cv is large, Cs slightly smaller. The results of simulation daily runoff in flood reason shows that the model can better characterize statistical parameters of different section, describle month runoff during the flood reason and reasonable feature of flood peak. It is feasible to nonparametric disaggregation model for stochastic simulation of hydrology, and result is reasonable.(4) This thesis discussed the method of the different matrix decomposition. When model were implemented by computer, it involved matrix decomposition. Both Cholesky decomposition and Schur decomposition can be decomposed for positive definite matrices, but for no positive definite matrices, Schur decomposition can be used. Because of a few stations and short series, it's hard to determine and need to further study that which decomposition method for the stations which covariance matrix are positive. (5) Improved nonparametric disaggregation model. The results of simulation monthly runoff using improved nonparametric disaggregation model shows that the model maintains good statistical properties of the measured sequences on mean, square deviation, Cv, maximum, minimum, first-order autocorrelation coefficient and second-order one, yet Cs relatively poor. The model simulated values are similar to measured ones overcomes the first autocorrelation not consistent. The results of simulation daily runoff in flood reason show that the simulation effects are basically consistent for both nonparametric disaggregation model and improved nonparametric disaggregation model.