基于1DCNN-LSTM尾矿坝浸润线预测  

Prediction of Saturation Line of Tailing Dam based on 1DCNN-LSTM

作　　者：杨玉好 杨斌 胡军[1] 董文宇 金实 YANG Yuhao;YANG Bin;HU Jun;DONG Wenyu;JIN Shi(School of Civil Engineering,University of Science and Technology Liaoning,Anshan 114051,China;Qidashan Beneficiation Plant of Anshan Iron and Steel,Anshan 114043,China)

机构地区：[1]辽宁科技大学土木工程学院,辽宁鞍山114051 [2]鞍山钢铁集团有限公司齐大山选矿厂,辽宁鞍山114043

出　　处：《有色金属工程》2024年第7期138-146,共9页Nonferrous Metals Engineering

基　　金：辽宁省教育厅面上项目(LJKZ0322);辽宁科技大学校青年项目(2020QN10)。

摘　　要：准确预测浸润线位置变化对尾矿坝的稳定性和安全性至关重要,为充分挖掘浸润线数据提供的空间特征和时序信息,提出将一维卷积神经网络(1DCNN)和长短期记忆神经网络(LSTM)相结合方法预测浸润线。以辽宁省齐大山风水沟尾矿库主坝为例,使用历史浸润线、库水位、坝体内外部位移、干滩长度5个主要因素作为模型输入数据,预测未来1 d和未来3 d的浸润线位置。将1DCNN-LSTM模型与经典的LSTM和反向传播神经网络(BP)进行对比研究。结果表明,1DCNN-LSTM浸润线预测的决定系数(R^(2))均在0.9以上,未来1 d的浸润线预测误差均值绝对值为0.004 m,最大误差绝对值为0.06 m,未来3 d的浸润线预测误差均值绝对值为0.003 m,最大误差绝对值为0.065 m,优于经典模型。这为短期浸润线预测提供一定的参考依据。Accurately predicting changes in the saturation line is crucial for the stability and safety of tailings dams.In order to fully explore the spatial characteristics and temporal information provided by saturation line data this paper proposes a combined approach using one-dimensional convolutional neural network(1DCNN)and long short-term memory neural network(LSTM)to predict saturation line.Taking the main dam of Fengshuigou tailing pond in Qidashan,Liaoning Province as an example,the model utilizes five major factors,including historical saturation line,reservoir water level,internal and external displacements of the dam,dry beach length,as input data to predict the saturation line positions for the next one day and three days.The 1DCNN-LSTM model is compared with classical LSTM and backpropagation neural network(BP)models.The results show that the Coefficient of Determination(R^(2))of the 1DCNN-LSTM infiltration line prediction is above 0.9,the absolute value of the mean error of the prediction error of the saturation line in the next one day is 0.004 m,the absolute value of the maximum error is 0.06 m,and the absolute value of the mean prediction error of the saturation line in the next three days is 0.003 m,and the absolute value of the maximum error is 0.065 m,which is better than that of the classical model.This provides a certain reference for the prediction of short-term saturation line.

关 键 词：1DCNN网络 LSTM网络 浸润线 尾矿坝 预测 

分 类 号：TD76[矿业工程—矿井通风与安全]