Combination Forecasting Model And Applied Research On Core Rockfill Dam Deformation

Dam deformation is usually caused by the coupling effects of various factors, therefore, it is highly complicated and unpredictable. Its uncertainty mainly lies on both gray characteristic, which is caused by incomplete data and incomprehensive considerations, and fuzziness between dam deformation and many factors. The gray characteristic and fuzziness make it difficult to analyze dam deformation. Until now, there is no such model or formual that could describe the complete characteristics of dam defoemation.Dam deformation is serious threat of dam safety.Therefore, it is an urgent problem to be solved to establish a precious, reliable and widelt applicable model to predict the dam defoemation trend.Current studies on dam deformation can't reflect the complexity due to lacking of comprehensive consideration of its grey characteristics and fuzziness. To solve the problem, this paper connects the project of a southwest core rockfill dam with a series of research work. The main results are as follows:(1) Since there are many factors with various expressions affecting dam deformation, it will cause the “dimension disaster” using all variables as inputs. To solve the problem, this article adopted the correlation analysis to determine the main factors affecting dam deformation, which include filling, water level and time. Then, we used the principal component analysis to reduce the main variables and finally obtained two principal components.(2) Because current studies on dam deformation can't reflect the complexity due to lacking of comprehensive consideration of its grey characteristics and fuzziness, to solve the problem, this article proposes the method adaptive neuro fuzzy inference system optimized grey model(ANFIS-GM) to analyze and predict dam deformation. This model has the comprehensive capacity of self-organizing, self-learning, adaptive and fuzzy inference. Through analysis of core rockfill dam settlement, it is proved that this model has smaller error than grey model which only considers the gray characteristics of dam deformation. It is also proved that ANFIS-GM can deal with small samples.(3) Genetic Algorithm(GA) optimized Error Back Propagation Artificial Neural Network(BP) and overcame the shortcomings that BP model easily falls into local optimal value. Through the analysis of core rockfill dam settlement, it is proved that this model can only deal with big samples.(4) The combination model can make full use of effective information from its sub-models and usually shows higher accuracy as a consequence. Traditional combination model does not consider time effects, this article proposes a nonlinear combination model which combine the time factor and BP model. This article also makes a comparison for the minimum of prediction error between this new model and the traditional model and prove that the new combination model can predict dam deformation better.The following two statements have been shown through analysis of core rockfill dam settlement. Firstly, both of the two kinds of combination models are usually more accurate than the ANFIS-GM model and GA-BP model. The combination models are also more stable with the ability of dealing with both big and small samples. Secondly, the nonlinear combination model which combine the time factor and BP model show only 4.65 mm lower than the combination model who's objective function is the minimum of prediction error's sum of squares,when they deal with prediction problem with small time span. They do not show much difference in accuracy. But in the prediction problem with large time span, the new combination model shows 22.19 mm higher accuracy than that of the combination model whose objective function is the minimum of prediction error's sum of squares in accuracy.