Study Of Hydrology Time Series Based On Modern Analysis Technology

Time series analysis method plays important roles in many fields including hydrologic law research, hydrologic analogy as well as hydrologic forecasting and so on. The hydrologic phenomena are highly nonlinear and complex, which demand applying, developing and creating new theories from higher chromatography and wider view to solve problems, which are unable to explain or continuous to appear in water science research now a days. In this dissertation, several modern analysis technologies such as artificial neural network, wavelet analysis, support vector machine, self-memory model and chaotic theory were applied to hydrologic time series forecasting respectively. To sum up, main results of this study were listed as follows:(1) After analyzing the main factors affecting generalization function of neural network, and aiming at poor representativity of hydrologic time series samples, standardization method of transforming extreme was presented to deal with input sample of time series, which could make the input samples lie in the interpolation range of training samples even if they lie in the extension of training samples, so it improved generalization function of neural network and accelerates network convergence speed.(2) A kind of model for hydrology time series forecasting were proposed using weighted summation of major periods wavelet coefficients to forecast periodic components. Considering the relativity between wavelet coefficients of wavelet transformation and time series, the method was based on the qualitative forecasting of continuous wavelet transformation and wavelet variance, and it took the advantage of multi-resolution characteristic of wavelet analysis effectively.(3) It was presented that a dynamic retrieving differential equation mode of two-variable hydrologic time series considering one main cycle. Bi-directional difference method and least square method were applied to deduce the parameters of differential equation, and the computing equation was given. On the basis of the dynamic retrieving differential equation, a two-variable hydrologic self-memory model was established( 4 ) It was presented that a dynamic retrieving differential equation mode of multi-variable hydrologic time series. According to antecedent research experience, a computing method of the dynamic retrieving differential equation of multi-variable hydrologic time series was given. On the basis of the dynamic retrieving differential equation, a multi-variable hydrologic self-memory model was established.(5) BP network method, RBF network method, least square support vector machine regression method and self-memory model were combined with chaos theory, and hydrologic time series forecasting models of chaos BP network, chaos RBF network, chaos least square support vector machine regression, chaos self-memory are accordingly established, respectively. Thereinto, combining the least square support vector machine method with chaos theory to the forecast of hydrologic time series and the phase space dynamic retrieving mode were the innovation view points of this dissertation.