基于动态MIC优化TCN的混凝土坝变形预测模型研究

Research on Deformation Prediction Model of Concrete Dam Based on Dynamic MIC Optimized TCN

混凝土大坝变形是其安全运行的主要控制指标,对大坝变形的预测就显得尤其重要。为降低常规预测模型因未考虑环境量的时滞性所带来的误差,采用最大信息系数方法(MIC)计算水位影响因子和大坝位移的时滞关系,并确定最佳滞后时间。进一步,采用时间卷积网络(TCN)解决了混凝土重力坝的变形预测的高维非线性难题,并不断逼近坝体位移影响因子矩阵空间和位移目标向量之间的最佳映射关系,进而得到大坝变形的预测模型,并将其预测结果与常用的ARIMA模型进行对比。结果表明,MIC-TCN模型比ARIMA模型有更好的预测效果。

The deformation of concrete dam is the main control index for its safe operation,and the prediction of dam deformation is particularly important.In order to reduce the error caused by the time lag of environmental quantity not considered in the conventional prediction model,the time lag relationship between water level influence factor and dam displacement is calculated by the maximum information coefficient method(MIC),and the optimal time lag is determined.Furthermore,the high-dimensional nonlinear problem of deformation prediction of concrete gravity dam is solved by using time convolution network(TCN),and the optimal mapping relationship between dam displacement influence factor matrix space and displacement target vector is continuously approached.Then the prediction model of dam deformation is obtained,and its prediction results are compared with the commonly used ARIMA model.The results show that the MIC-TCN model has better prediction effect than ARIMA model.