Study On The Forecasting Of River Runoff

The water resources system is a large-scale complicated system. Forecasting the water resources system accurately and timely has the important significance in the theories and the projects. Runoff in the water resources is affected by many factors such as physical geography environment of drainage area and human activities. And its change characteristics and rules are intricate putting up random, grey, nonlinear characteristics and so on.We study on the forecasting of river runoff in the paper. There are four methods in forecasting the different characteristics of annual runoff.In the paper wavelet analysis method is used for forecasting the whole change tendency of annual runoff. The wavelet analysis has the time-frequency local change characteristic. We can use the "focusing" character to reveal the fine structure of hydrology water resources time series. The method provides a new way in analyzing multiple time-scale change and the distribution and the discontinuity points. The results from Marr wavelet function not only unfolds the hydrology water resources time series' frequency characteristic in the time domain, but also give intensity of each kind of time-scale (in one cycle) clearly and distribution situation as well as discontinuity point, moreover, it can analyze its main cycle.In the paper Markov-chain forecasting theory especially weighted Markov-chain forecasting theory is used to forecast the annual runoff in abundant and dried condition. Weighted Markov-chain forecasting theory can achieve the goal of fully and reasonably using the information to forecast the runoff. Though there are vacancy data in four years, we use Newton interpolation to complete the data. Then, we get the better forecasting result by weighted Markov-chain forecasting theory, which demonstrates that weighted Markov-chain forecasting theory uses the data information fully and reasonably.In the paper two methods are used for forecasting the annual runoff, they are time-series analysis and artificial neural networks. ARIMA model and BP networks with three layers are mainly used in the paper. ARIMA model mainly uses data relevance to build the model. ARIMA (3,2,1) built by analysis is the best model, while the results are not good. Therefore, in order to get good forecasting result we combine two ARIMA models to build the new model, then we get the better forecasting result. BP network with three layers mainly uses L-M method to build the hydrologic forecasting model . This method enables the forecasting precision to achieve the national hydrologic forecasting standard.In the paper Matlab software and Eviews software is used for processing the data.This paper uses many models. In each chapter each model building process is introduced specifically. Conventional models are already quite mature in runoff forecasting in application, so in the paper we stress the study on the combination between improving models and relevant models.