摘要
Abstract
In the prediction of dam deformation,the chaotic characteristics and long-range dependencies hidden in the displacement time series data of measuring points enhance the difficulty of capturing spatial correlation and multi-scale deformation mechanisms.Pure spatial convolution models are difficult to analyze the dynamic evolution laws of time series,which can easily lead to the accumulation and propagation of prediction errors.Therefore,this study proposes a deformation prediction method for reinforced earth dams with concrete impermeable walls in hydraulic engineering based on Convolutional Neural Network Bidirectional Long Short Term Memory(CNN BiLSTM).Firstly,using the Hierarchical Agglomerative Clustering(HAC)algorithm,a similarity matrix of measurement points is constructed based on Euclidean distance,and the Ward variance minimization criterion is used to merge the clusters with the smallest distance layer by layer,adaptively identifying spatial regions with similar deformation patterns.Then,the partitioned data is reconstructed into a two-dimensional tensor,and local spatial features are extracted using the local receptive field and weight sharing mechanism of the convolutional layer.The feature map is compressed using global average pooling operation to extract spatial features of the dam monitoring data.Finally,the extracted global feature vectors are concatenated with time-series data such as displacement and water pressure at measurement points to construct an input sequence.The BiLSTM bidirectional loop structure is used to synchronously capture long-term dependencies and short-term fluctuation features.Its gating mechanism dynamically adjusts the weight allocation between historical information and current input,effectively suppressing noise interference in chaotic time-series.Combining historical deformation trends with contextual information,use a fully connected layer to output the predicted deformation value for the next moment.The test results show that when using the proposed method to predict the deformation of reinforced earth dams with impermeable walls,the average closeness of the deformation curve is 0.94,which has a relatively ideal prediction effect.关键词
水利工程/混凝土防渗墙/加固土坝/变形预测Key words
water conservancy engineering/concrete waterproof wall/reinforced earth dam/deformation prediction分类
信息技术与安全科学
