The long‐term safety and health monitoring of large dams has attracted increasing attention. In this paper, coupling prediction models based on long short‐term memory (LSTM) network are proposed for the long‐term deformation of arch dams. Principal component analysis (PCA) and moving average (MA) method, adopted to make dimension reduction for the input variables, are respectively combined with the LSTM to achieve two coupling prediction models, that is, LSTM‐PCA and LSTM‐MA. Lijiaxia arch dam, which has been in operation over 20 years, is taken as an analysis example. Compared with the traditional hydrostatic‐seasonal‐time model, the hydrostatic‐seasonal‐time thermal model, and the multilayer perceptron model, the proposed models show more effectiveness concerning the predicted displacements of the arch dam. The accuracy of the predicted results from the coupling prediction models is better. Furthermore, the coupling prediction models could capture the long‐term characteristics and provide better prediction with short monitoring data. Compared with the LSTM‐PCA model, the LSTM‐MA model is more suitable for engineering applications due to its convenience.

中文翻译：

---

基于长短期记忆网络的拱坝长期位移耦合预测模型

大型水坝的长期安全和健康监控已引起越来越多的关注。本文针对拱坝的长期变形，提出了基于长短期记忆（LSTM）网络的耦合预测模型。为了减少输入变量的尺寸而采用的主成分分析（PCA）和移动平均（MA）方法分别与LSTM结合使用，以实现两个耦合预测模型，即LSTM-PCA和LSTM-MA。以已经运行了20年的李家峡拱坝为例进行分析。与传统的静水-季节-时间模型，静水-季节-时间热模型和多层感知器模型相比，所提出的模型在预测拱坝位移方面表现出更大的有效性。来自耦合预测模型的预测结果的准确性更好。此外，耦合预测模型可以捕获长期特征，并以较短的监测数据提供更好的预测。与LSTM-PCA模型相比，LSTM-MA模型具有便利性，因此更适合工程应用。