基于SSA-LSTM的土石坝浸润线预测模型  

SSA-LSTM-Based Prediction Model for the InfiltrationLine of Earth and Rock Dams

作　　者：刘振宇 傅蜀燕 赵定柱 王奎 欧斌 LIU Zhen-yu;FU Shu-yan;ZHAO Ding-zhu;WANG Kui;OU Bin(College of Water Conservancy,Yunnan Agricultural University,Kunming 620201,China;The National Key Laboratory of Water Disaster Prevention,Nanjing 210098,China;Yunnan Province Small and Medium-sized Water Conservancy Engineering Research Centre for Intelligent Management and Maintenance,Kunming 650201,China)

机构地区：[1]云南农业大学水利学院,昆明650201 [2]水灾害防御全国重点实验室,南京210098 [3]云南省中小型水利工程智慧管养工程研究中心,昆明650201

出　　处：《科学技术与工程》2024年第34期14813-14820,共8页Science Technology and Engineering

基　　金：国家自然科学基金(52069029,52369026);水灾害防御全国重点实验室2023年度“一带一路”水与可持续发展科技基金项目(2023490411);云南省农业基础研究联合专项面上项目(202401BD070001-071);云南省水利水电工程安全重点实验室开放课题基金(202302AN360003)。

摘　　要：浸润线的异常分布对土石坝的稳定性和安全性构成严重威胁。因此,准确预测浸润线成为土石坝安全监控的核心任务。然而,测压管监测数据存在非线性和非平稳性问题,这使得浸润线的预测模型容易出现过拟合,进而影响预测精度。为了解决这一问题,提出了一种基于SSA-LSTM模型的浸润线预测方法。该方法结合了麻雀搜索算法(sparrow search algorithm, SSA)和长短期神经网络(long-short-term neural Network, LSTM),通过优化模型的初始学习率和正则化参数,使输入数据与网络结构更好地匹配,从而提高预测精度。通过决定系数(R^(2))、平均绝对误差(mean absolute error, MAE)和均方根误差(root mean square error, RMSE)三个定量评价指标对模型预测结果进行了评估。结果表明,与传统的LSTM等模型相比,SSA-LSTM模型的预测精度有了显著提高,为土石坝的浸润线预测提供了有价值的参考。The abnormal distribution of infiltration lines poses a serious threat to the stability and safety of earth and rock dams.Therefore,accurate prediction of the infiltration line has become a core task in the safety monitoring of earth and rock dams.However,there are nonlinear and non-smoothness problems in the monitoring data of the pressure tube,which makes the prediction model of the dip line prone to overfitting,which in turn affects the prediction accuracy.In order to solve this problem,a dip line prediction method based on SSA-LSTM model was proposed.The method combines SSA(sparrow search algorithm)and LSTM(long-short-term neural network)to improve the prediction accuracy by optimizing the initial learning rate and regularization parameters of the model to better match the input data with the network structure.The model prediction results were evaluated by three quantitative evaluation metrics,namely,coefficient of determination(R^(2)),MAE(mean absolute error)and RMSE(root mean square error).The results show that the prediction accuracy of the SSA-LSTM model is significantly improved compared with the traditional LSTM and other models,which provides a valuable reference for the prediction of the dip line of earth and rock dams.

关 键 词：土石坝 预测模型 浸润线 麻雀搜索法 长短期神经网络 

分 类 号：TV698.1[水利工程—水利水电工程]