基于优化的EMD-LSTM的土石坝沉降预测模型研究

The study on settlement prediction model of earth-rock dams based on EMD-LSTM

摘要

针对土石坝沉降预测模型中回归模型易受多重共线性影响，神经网络模型存在过拟合、局部极值陷阱以及超参数难以确定等问题，提出了一种基于经验模态分解(EMD)和长短期记忆神经网络(LSTM)的优化模型。首先，通过EMD对全球导航卫星系统(GNSS)测点的时间序列数据进行多尺度分解，提取趋势和周期成分。然后，利用主成分分析(PCA)筛选关键影响因子，减少数据维度，提高模型的泛化能力。最后，采用LSTM构建时间序列模型，并通过鲸鱼优化算法(WOA)优化LSTM的超参数，以提升模型的预测精度和收敛速度。实验结果表明，该模型在土石坝沉降预测中具有显著的优势，均方误差(MSE)为7.070 1,平均绝对误差(MAE)为1.885 9,拟合优度(R²)为99.83%。与传统方法相比，该模型在降噪、特征捕捉和超参数优化等方面均有明显提升，可为土石坝沉降提供可靠的预测方案。

Abstract

In response to the challenges faced by settlement prediction models for earth-rock dams, such as the susceptibility of regression models to multicollinearity, and issues like overfitting, local minima traps, and difficulty in determining hyperparameters in neural network models, an optimized model was proposed based on Empirical Mode Decomposition(EMD) and Long Short-Term Memory(LSTM) neural networks. Firstly, EMD is employed to perform multi-scale decomposition of time series data from Global Navigation Satellite System(GNSS) measurement points, extracting trend and periodic components. Then, Principal Component Analysis(PCA) is utilized to select key influencing factors, reducing data dimensionality and enhancing the generalization capability of the model. Finally, an LSTM is used to construct the time series model, and the Whale Optimization Algorithm(WOA) is applied to optimize the hyperparameters of the LSTM, improving the model′s prediction accuracy and convergence speed. The experimental result show that this model offers significant advantages in the settlement prediction of earth-rock dams, with a Mean Squared Error(MSE) of 7.070 1, a Mean Absolute Error(MAE) of 1.885 9, and a coefficient of determination(R²) of 99.83%. Compared to traditional method, this model demonstrates notable improvements in noise reduction, feature capture, and hyperparameter optimization, providing an accurate and reliable solution for settlement prediction in earth-rock dams.

关键词

土石坝 / 沉降预测 / 模型 / 经验模态分解(EMD) / 长短期记忆神经网络(LSTM)

Key words

earth-rock dam / settlement prediction / model / empirical mode decomposition(EMD) / long short-term memory(LSTM)

Author summay

李宗淇(1999—), 男, 硕士研究生, 主要从事水工结构工程研究。E-mail: [e-mail]

基金资助

中国水利水电科学研究院基本科研业务费专项项目(ZS0145B012024)