摘要
尾矿库作为一种危险源,需对其进行长期监测及预警,但目前针对尾矿库的监测方法费时费力且预警模型较少,为此提出了一种基于小基线集(SBAS-InSAR)技术、长短期记忆(LSTM)神经网络以及网格搜索(GS)算法相结合的尾矿库沉降预测模型,实现了尾矿库沉降监测与预测的一体化。首先,以60景Sentinel-1A影像作为数据源,采用SBAS-InSAR技术监测鞍山市西果园尾矿库动态沉降过程,获取该尾矿库2018—2020年内时间序列沉降信息,将其与GPS技术获取的测量结果进行对比,发现时序In SAR监测精度较高。然后将降雨量与沉降量关联分析,得到尾矿库沉降的波动规律,构建LSTM神经网络沉降预测模型,再利用GS算法将模型中的超参数进行全局寻优。最后将监测数据划分为训练集与测试集,与传统预测模型进行对比。试验结果表明:GS-LSTM模型在西果园尾矿库沉降预测中呈现出了较好的预测精度,3个测试点中最大平均绝对误差(MAE)为2.51 mm,最大均方根误差(RMSE)为2.90 mm,可以较为精准地反映出具有尾矿库沉降特点的波动和趋势,为尾矿库灾害预警及治理提供了理论依据。
As a source of danger,tailings ponds need to be monitored and warned for a long time.However,monitoring methods for tailings ponds are time-consuming and labor-intensive at present,and warning models are few.Therefore,a prediction model for tailings ponds settlement based on small baseline set(SBAS-InSAR)technique,long term and short term memory(LSTM)neural network and grid search(GS)algorithm is proposed.The integration of settlement monitoring and prediction of tailings pond is realized.Firstly,with 60 scenes Sentinel-1A image as the data source,SBAS-InSAR technique is used to monitor the dynamic settlement process of Anshan Xiguoyuan Tailings Pond,obtaining the time series settlement information of the tailings pond from 2018 to 2020,and comparing it with the measurement results obtained by GPS technique.It is found that the time series In SAR monitoring accuracy is high.Then,the correlation analysis of rainfall and settlement is carried out to obtain the fluctuation law of tailings pond settlement.The LSTM neural network settlement prediction model is constructed,and the GS algorithm is adopted to globally optimize the hyperparameters in the model.Finally,the monitoring data are divided into training set and test set,and compared with the traditional prediction model.The experimental results show that the GS-LSTM model shows good prediction accuracy in the settlement prediction of Xiguoyuan Tailings Pond.The maximum mean absolute error(MAE)of the three test points is 2.51 mm,and the maximum root mean square error(RMSE)is 2.90 mm.It can accurately reflect the fluctuations and trends with the characteristics of tailing pond settlement,providing a theoretical basis for disaster early warning and governance of tailings pond.
作者
李如仁
孙加瑶
LI Ruren;SUN Jiayao(School of Transportation and Geomatics Engineering,Shenyang Jianzhu University,Shenyang 110168,China)
出处
《金属矿山》
CAS
北大核心
2023年第1期102-109,共8页
Metal Mine
基金
国家自然科学基金项目(编号:51774204)。