A Study Of Chaotic Characteristics Of Tumen River Runoff Time Series And The Comparative Study Of Forecasting Methods

Floods are one of the most serious disasters in our country. And runoff forecast is not only an important part of flood prevention but also strong measures of rational use of hydropower and water resources. Tumen River Basin floods sometimes. And the accurate prediction of runoff can reduce damage. Chaos Theory has been a nonlinear science with nearly half a century development. A new area of research arises after Combining Chaos Theory with hydrological science. It can reproduce the dynamic characteristics of the system to reconstruct time series of chaotic characteristics with phase space reconstruction theory. It can get very good prediction accuracy on the basis of the forecast. The purpose of the study is the runoff data measured monthly by hydrological stations in the main stream areas of Tumen River. We will recognize the monthly runoff time series of chaotic characteristics. We will use a variety of methods for chaotic prediction of time series of monthly runoff. By comparing the prediction, we can get the best chaotic prediction method.First of all, the chaos identification and prediction are carried out on the basis of the phase space reconstruction. This paper determines the time delay by the autocorrelation function method and mutual information of the phase space reconstruction method. it can determine delay time of three hydrological station of the Tumen river mainstream. Using saturated correlation dimension method and the Cao method determine the embedding dimension in reconstructive space. Based on this judge, whether it has the characteristics of chaos time series. And using Lyapunov method and the maximal Lyapunov exponent method to identify chaos characteristics. And the results show that the time series of hydrological station of the Tumen river runoff has chaotic character.Second, we will use the weighted one-rank local-region method on the basis of space reconstruction, which is on the basis of the maximal Lyapunov exponent method. We will have a forecast by Volterra adaptive filter prediction mode and RBF neural network model. Comparing these results, RBF neural network model is the best method to forecast runoff in Tumen River. Overall various methods on runoff of the trend forecast are accurate, have certain prediction ability.Finally, the paper summarized the research work, and puts forward the research direction in the future. It provides scientific basis not only for making the strategy of developing water resources but also effective guidance of disaster prevention and disaster-relief.