融合动态图卷积与时序卷积的多序列渗流压力预测方法研究OA

The evolution of seepage conditions is directly related to the long-term operation safety of earth-rock dams.To enhance the perception and early warning capability of seepage pressure trends,a multi-sequence seepage pressure prediction method that integrates dynamic graph convolution and temporal convolution is proposed.A sliding correlation mechanism is used to construct a dynamic adjacency matrix,capturing time-varying spatial dependencies among monitoring points,and the graph convolutional network(GCN)is employed to extract structural features,while the temporal convolutional network(TCN)is introduced to capture long-term dependencies,thereby enabling accurate prediction of seepage trends.Based on multi-year measured seepage pressure data from a large earth-rock dam in Southwest China,a series of comparative experiments are conducted.The results show that incorporating dynamic graph structures can improve model performance by approximately 18%,and when replacing the TCN with a multilayer perceptron(MLP),the MAE increased to 1.26 and the MAPE rose to 9.59%,confirming the critical role of TCN in capturing temporal dependencies.