Application Of Analog Complexing Algorithm In Runoff Forecasting

Hydrological forecasting is an important part of hydrology. It is very useful in flood control, drought relief and many other fields. With the development of national economy and the exploitation of water resource, hydrological forecasting is becoming more and more important. A nonparametric regression method, Analog Complexing (AC) algorithm is used for runoff forecasting in arid regions. The paper consists of the following parts: the analysis of chaotic characteristics in runoff time series, the application of AC in runoff forecasting, the combination of parametric regression and nonparametric AC and it's application in runoff forecasting.In the first part, the chaotic characteristics of annual and monthly runoff of the source streams of Tarim River are analyzed. The paper applies popular and effective methods, such as saturated correlation dimension method and the largest Lyapunov exponent method, to analyze the chaotic characteristics of these hydrological time series. The results show there is more or less chaos in these hydrological time series.In the second part, the AC is used for the prediction of annual runoff of Yarkant River and Hotan River. Heuristic test is employed in the modeling process to determine the best pattern length, the method of pattern similarity measures and the number of analogues selected. The relation between the best pattern length and the largest Lyapunov exponent is analyzed. The comparison between measured and predicted runoff shows this prediction model works well.In the third part, the same method is used to develop prediction models for the original monthly runoff, runoff of each month and standardized monthly runoff time series, and the results of forecasting are compared with the measured runoff. The process of modeling validates the relation between the best pattern length and the largest Lyapunov exponent. The prediction results with the standardized runoff series are better than others.In the last part, the parametric method is combined with the nonparametric method AC and is applied to the prediction of standardized monthly runoff. The regression parameters are obtained from partial least square regression method. It is proved to be effective in runoff forecasting.