The Study Of Mid-long Term Runoff Forecasting Model On Three Gorges Cascade And The Design And Application Of The System

The formation of runoff is subject to many factors such as precipitation, evaporation,runoff yield and concentration, terrain, physiognomy and human activities. It is markedwith complicated non-linearity. The mid-long term runoff forecasting is beneficial to thedecision-making of hydro-junction as soon as possible and the overall planning of optimalcollocation of cascade water resource because of its long forecast period. However, asthere are many ever-changing factors affecting the forecast, the mid-long term runoffforecasting is not suitable to the short-term runoff forecasting. Nevertheless, our country isfrequently stricken with natural disasters, so how to increase the precision of the mid-longterm runoff forecasting is a problem to deal with.As an important civilian-related project, The Project of Three Gorges plays asignificant role in many areas like hydropower and shipping, especially in flood-control.This paper focuses on the study of the structure and function of the Mid-long Term RunoffForecasting system on Three Gorges Cascade by designing a hydro-database and bringingforowrd a forecasting model of variable step size BP neural network based on principalcomponent analysis, utilizing principal component analysis to remove the relativity ofinput data and decrease the input dimension, which effectively lower the modelcomplexity. After in-depth study, this paper puts forward a selective neural networkensemble model based on principal component analysis, which constructs multiplysubnets through different classifiers, and apply to the sample of every subnet the principalcomponent analysis to remove the relativity of input data and decrease the inputdimension. Meanwhile, it uses the bagging method to deal with the sample to increase thediscrepancy and integrates different results from different classifiers with differentselecting ways, raising study efficiency and generalization ability of the model. It refers tothe average runoff material from Yichang Station of The Three Gorges cascade to predictand verify the model and compare this model with the traditional BP neural networkmodel. The model and algorithm will be integrated into the Mid-long Term RunoffForecasting system on Three Gorges Cascade at last. The results show that the improvedmodel overshadows the traditional one in the area of operating efficiency and forecastprecision, which paves the way for mid-long term runoff forecasting and is promising inproject application and worth popularizing.This study is also applied in the subject of national science and technology support plan named “the combined scheduling technology of The Three Gorges and yangtzeUpstream oversize cascade hub group”.