Research On Mid-long Term Runoff Forecasting Under Water And Soil Conservation Measure

The runoff was influenced deeply by human activities in recent years. How to accuratelyforecast the the runoff under water conservancy is the key problem which need to be solvedby scientific and technical workers. This article as Tuweihe watershed where kinds of soil andwater conservation measures were implemented on early an study area to analysis the rainfalland runoff timing change characteristics, reveals runoff evolution, and recognition drivingforce of runoff from statistics method and causes analysis, define quantitative index toquantitative analysis the influence by soil and water conservation measures.From the angle ofthe influence of water cycle to establish mid-long term runoff forecast model based on thesoil and water conservation measures with the artificial neural network, the support vectormachine and the gray theory method, to provide the theory basis for the rational developmentand utilization of water resources. The mainly contents and conclusions are as follows:(1) Analysised the characteristics of runoff in the interannual and years distribution. Therunoff sequence changed in1978years, the variation declined but well-distributed.Using themathematical statistics and cause analysis methods to recognition the driving force. And thinkof the runoff is mainly affected by soil and water conservation measures with the similarclimate conditions. Take average runoff coefficient as a quantitative index to quantificationinfluences soil and water conservation measures on runoff amount in different period. As1956~1978years for the baseline period,1979~2005years average annual runoff is reduced17.4%similar to annual precipitation conditions of the baseline period, and similar in floodseason precipitation conditions average runoff reduced22.5%;To establish a mathematicalstatistical model between the soil and water conservation measures in the area of growth rateand decrease the rate of runoff coefficient.(2) Introduced the basic principle, modeling steps and evaluation rules of the BP neuralnetwork and RBF neural network, and adopting the two models to predict annual runoff andflood runoff under water and soil conservation measures influeced. Compared thecharacteristics and accuracy of BP model and RBF model,think of the RBF neural networkmodel is suitable to forecast annual runoff,and the BP neural network model is suitable toforecast flood runoff.(3) Introduced support vector machine (SVM) and the least square support vector machine (LSSVM) which has a good generalization performance and the strong ability to dealwith the nonlinear problems.Using the above two kinds methods establish the mid-long termrunoff forecast model, think of support vector machine forecasting model is applied to mid-long term runoff forecast is effective, and the SVM model is better than the LSSVM model.(4) Considering the runoff sequence has the downward trend, According to the GM(1,1)model's prediction trend performance and SVM model's nonlinear processing powerestablished the GM (1,1) SVMcombination model. Introduced the mathematical principleand algorithm of gray prediction model, put forward the basic ideas and modeling steps, andthe combination model is applied to the annual runoff and flood runoff forecasting. Annualrunoff forecast effect is good, but the flood runoff forecast effect unsatisfactory, needs furtherresearch.(5) The features of various models, precision and practicability are compared,recommend the SVM model as the suitatable runoff forecasting model in the practicalapplication at Tuweihe watershed, put forward to use the similar to design flood frequencyamplification method for scaling principle of predicting annual runoff distribution process ofthe new thinking.