基于长短期记忆网络的黄河小浪底水库高斜心墙堆石坝沉降预测模型及其预测精度分析  

Settlement Prediction Model of High Inclined Rockfill Dam and Its Prediction Accuracy Analysis Based on LSTM

作　　者：王亚坤 傅志敏[1] 苏正洋 WANG Ya-kun;FU Zhi-min;SU Zheng-yang(College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China;Nanjing Hydraulic Research Institute,Nanjing 210029,China)

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [2]南京水利科学研究院,江苏南京210029

出　　处：《水电能源科学》2022年第2期110-113,共4页Water Resources and Power

基　　金：国家重点研发计划(2018YFC0407105);国家自然科学基金项目(51779154,51979176)。

摘　　要：针对黄河小浪底水库高斜心墙堆石坝沉降预测精度低,无法准确反映大坝运行性态等问题,基于黄河小浪底水利枢纽主坝沉降监测数据,引入长短期记忆网络(LSTM)方法建立高斜心墙堆石坝沉降预测模型对主坝沉降进行预测,并对不同损失函数下模型优化过程进行分析,最终选定平均绝对误差作为损失函数。预测结果表明,LSTM方法各项指标结果均优于随机森林算法和BP神经网络算法,可实现高精度预测大坝运行性态,具有较高实用性,可为水库大坝沉降预测及工程运行管理提供借鉴和参考。To solve the problem that the settlement prediction accuracy of high inclined core rockfill dam is low and cannot reflect the exact operation behavior of the dam, LSTM was introduced to establish the settlement prediction model of high inclined core rockfill dam based on the settlement monitoring data of the main dam of Xiaolangdi Hydropower Project of the Yellow River. The optimization process of the model was analyzed for different loss function, and the mean absolute error was chosen as the loss function. The prediction results show that each index of the LSTM is better than stochastic forest algorithm and BP neural network algorithm, which realizes high-precision prediction of operation behavior of high-inclined core rockfill dam, has high practicability. Thus, it provides a guarantee for prediction of reservoir dam settlement and operation and management of the project.

关 键 词：神经网络 土石坝沉降 预测 长短期记忆网络 

分 类 号：TV641[水利工程—水利水电工程] TV698.11