基于IHHO-LSTM-KAN的大坝变形预测模型

丁勇康; 远近; 毛延翩; 都旭煌; 齐智勇; 苏怀智

doi:10.13928/j.cnki.wrahe.2025.05.014

水利水电技术（中英文） ›› 2025, Vol. 56 ›› Issue (5) : 170 -182. DOI: 10.13928/j.cnki.wrahe.2025.05.014

水工建筑

基于IHHO-LSTM-KAN的大坝变形预测模型

丁勇康 ¹^,² ,
远近 ³ ,
毛延翩 ³ ,
都旭煌 ³ ,
齐智勇 ³ ,
苏怀智 ¹^,²^,⁴

作者信息 +

Dam deformation prediction model based on IHHO-LSTM-KAN

Yongkang DING ¹^,² ,
Jin YUAN ³ ,
Yanpian MAO ³ ,
Xuhuang DU ³ ,
Zhiyong QI ³ ,
Huaizhi SU ¹^,²^,⁴

Author information +

文章历史 +

PDF

摘要

【目的】全生命周期高精度的变形预测是评估大坝服役性态和保障大坝安全运行的关键方法。目前预测模型存在数据特征相关性解析不足、对短时序数据预测精度不高、忽视时序持续增长的特性、模型训练易陷入局部最优等问题。【方法】提出一种大坝变形预测模型，利用长短期记忆网络(LSTM)捕捉时序长短期依赖关系，并耦合KAN机制改进网络全连接层结构以增强对长短时序复杂数据关系的表征能力，采用多策略改进的哈里斯鹰优化算法(IHHO)探索超参数最优组合，从而优化模型结构、解决梯度问题、加速训练收敛并提高预测性能。【结果】实例表明，该模型对长短时序的预测精度和泛化能力均优于其他深度学习模型，收敛速度优于其他智能优化算法，KAN机制对短时序预测的改进效果较为明显。【结论】所建模型具有较好的稳健性与适用性，可为大坝全生命周期的安全监测提供技术参考。

Abstract

[Objective] High-precision deformation prediction during the whole life cycle is a key method to evaluate the service behavior of dams and ensure the safe operation of dams. The current prediction model has problems such as insufficient correlation analysis of data feature, low prediction accuracy of short time series data, neglecting the continuous growth properties of the time series, and easy to fall into the local optimum in model training. [Methods] Therefore, a dam dynamic deformation prediction model is proposed, which utilizes the long short-term memory neural network(LSTM) to capture the long-term and short-term dependence of time series, couples the Kolmogorov-Arnold Networks(KAN) mechanism to improve the fully connected layer structure of the network to enhance the ability to characterize the complex data relationship of long and short time series, and adopts multi-strategy improved Harris Hawks optimization algorithm(IHHO) to explore the optimal combination of hyperparameters, so as to optimize the model structure, solve the gradient problem, accelerate the training convergence and improve the prediction performance. [Results] Examples show that the prediction accuracy and generalization ability of the model for short and long time series are better than other deep learning models, and the convergence speed is superior to other intelligent optimization algorithms, and the improvement effect of KAN mechanism on the short time series prediction is more obvious. [Conclusion] The model has good robustness and applicability, which can provide technical reference for the dynamic safety monitoring of the whole life cycle of dams.