中国农村水利水电
中國農村水利水電
중국농촌수이수전
China Rural Water and Hydropower
2015年
9期
139-143
,共5页
陈勋辉%程中凯%景继%黄耀英%万智勇
陳勛輝%程中凱%景繼%黃耀英%萬智勇
진훈휘%정중개%경계%황요영%만지용
优化反演%均匀设计%BP神经网络%水压分量
優化反縯%均勻設計%BP神經網絡%水壓分量
우화반연%균균설계%BP신경망락%수압분량
optimization inversion method%uniform design%BP neural networks%water pressure component
由于高拱坝坝体和基岩工作条件复杂，难以准确给定计算参数，因此，基于实测变形反演高拱坝和基岩力学参数。首先采用均匀设计原理对坝体和基岩综合弹性模量进行设计参数组合，通过有限元正分析得到多个典型测点处坝体的水压分量差值样本，以水压荷载作用下的水压分量差值作为神经网络的输入，对应的待反演参数的样本值作为输出，训练神经网络建立描述待反演参数同坝体多个典型测点变形的非线性关系。然后，将实测分离出的各典型测点水压分量的差值输入到训练好的神经网络，即可得到各参数的反演值。以西南某高拱坝为例，反演分析了坝体以及坝基主要岩体的弹性模量等参数。结果表明：该方法下各典型测点水压荷载下变形计算差值同实测变形水压分量差值比较接近，可知反演得到的参数较合理，以上反演方法是可行的。
由于高拱壩壩體和基巖工作條件複雜，難以準確給定計算參數，因此，基于實測變形反縯高拱壩和基巖力學參數。首先採用均勻設計原理對壩體和基巖綜閤彈性模量進行設計參數組閤，通過有限元正分析得到多箇典型測點處壩體的水壓分量差值樣本，以水壓荷載作用下的水壓分量差值作為神經網絡的輸入，對應的待反縯參數的樣本值作為輸齣，訓練神經網絡建立描述待反縯參數同壩體多箇典型測點變形的非線性關繫。然後，將實測分離齣的各典型測點水壓分量的差值輸入到訓練好的神經網絡，即可得到各參數的反縯值。以西南某高拱壩為例，反縯分析瞭壩體以及壩基主要巖體的彈性模量等參數。結果錶明：該方法下各典型測點水壓荷載下變形計算差值同實測變形水壓分量差值比較接近，可知反縯得到的參數較閤理，以上反縯方法是可行的。
유우고공패패체화기암공작조건복잡，난이준학급정계산삼수，인차，기우실측변형반연고공패화기암역학삼수。수선채용균균설계원리대패체화기암종합탄성모량진행설계삼수조합，통과유한원정분석득도다개전형측점처패체적수압분량차치양본，이수압하재작용하적수압분량차치작위신경망락적수입，대응적대반연삼수적양본치작위수출，훈련신경망락건립묘술대반연삼수동패체다개전형측점변형적비선성관계。연후，장실측분리출적각전형측점수압분량적차치수입도훈련호적신경망락，즉가득도각삼수적반연치。이서남모고공패위례，반연분석료패체이급패기주요암체적탄성모량등삼수。결과표명：해방법하각전형측점수압하재하변형계산차치동실측변형수압분량차치비교접근，가지반연득도적삼수교합리，이상반연방법시가행적。
As the complexity and uncertainty of arch dam body and bedrock ,it is very difficult to obtain calculated parameters accu‐rately .As a result ,the mechanical parameters of high arch dams based on measured deformation is back-analyzed in this paper . Firstly ,by combining the design parameters of comprehensive elastic modulus of dam body and bedrock through uniform design theo‐ries by using the finite element method to obtain the sample of the difference value of water pressure component separated from sever‐al typical points ,then the difference value of water pressure component under water pressure is taken as the import of neural net‐works ,the unknown inverse parameters are taken as outputs correspondingly ,the neural network was established .It was trained to describe the nonlinear relationship between inverse parameters and dam displacements of several typical points .Secondly ,the actual difference value of water pressure component separated from several typical points is input into the trained neural networks to obtain the real material parameters .As an example ,the elastic modulus of dam bodies and rock masses of dam foundations under the base of a high arch dam in southwest are back-analyzed by the above method .It indicates that the difference values of water pressure com‐ponent of several typical points under water pressure were almost the same as the actual difference values ,so it can be concluded that the inversion values are reasonable ,and this inversion method is feasible .
