HDNN-MC模型在特高拱坝变形预测中的应用

Engineering application of HDNN-Markov model in high arch dam deformation forecast

递阶对角神经网络(HDNN)采用动态BP学习算法,可以逼近任意非线性函数且具有收敛速度快、预报精度高的特点,因此本文将其引入到大坝安全监测领域,以水压、温度和时效为输入量,坝体位移为输出量,在此基础上运用马尔科夫链(MC)模型对预测数据进行残差计算和状态划分,确定马尔科夫链状态概率矩阵,通过马尔科夫链状态概率矩阵对HDNN模型进行反馈修正,从而提高精度。基于此建立了HDNN-MC模型并应用于某特高拱坝的变形预测。结果表明,HDNN-MC综合模型相对于单一模型,预测精度得到显著提高,能更高效准确地预测大坝变形。

Hierarchical Diagonal neural network uses dynamic BP learning algorithm to modify connection weights,such that it can approximate any nonlinear function and have the characteristics of convergent quickly and high forecast accuracy.In this pa-per,through residual error analysis and state division,the state transferring probability matrix is determined,and the HDNN model is fed back by adj usting the state transferring probability matrix to improve prediction accuracy.A prediction method based on HDNN-Markov model is established and applied to forecast the deformation of a high arch dam.As compared to the measured values,the predicted results indicate that the HDNN-Markov model has higher prediction accuracy and better practicability.