水电能源科学

鉴于大坝变形的准确预测对确保工程安全运行至关重要，提出了一种基于改进完全自适应噪声集合经验模态分解(ICEEMDAN)与TCN-KAN组合模型的大坝变形预测方法。首先，对大坝变形实测数据进行精细化分解，通过模糊熵分析筛选和重构模态分量，将变形实测序列划分为重构分量和随机分量；其次，利用时间卷积神经网络与Kolmogorov-Arnold网络模型分别进行分量预测。应用实例表明，所提模型在预测精度上具有显著优势，为大坝变形性态的预测提供了一种有效方法。

Accurate prediction of dam deformation is crucial for ensuring the safe operation of dams. This paper proposes a dam deformation prediction method based on the combination of ICEEMDAN and TCN-KAN. Firstly, the measured dam deformation data is finely decomposed, and modal components are selected and reconstructed using Fuzzy entropy analysis. The original deformation sequence is divided into reconstructed deformation and random deformation. Then, temporal convolutional network and Kolmogorov-Arnold network model are employed to predict these two components separately. Case studies show that the proposed model has significant advantages in prediction accuracy, which provides an effective method for forecasting dam deformation patterns.