形变监测与预测是对水电站异常情况进行预警和及时采取补救措施的关键。本文提出了一种长短期记忆(LSTM, Long Short-Term Memory)神经网络方法来预测大渡河流域瀑布沟水电站干涉合成孔径雷达(InSAR, Interferometric Synthetic Aperture Radar)的时间序列形变。该方法首先利用多时相干涉合成孔径雷达(MT-InSAR, Multi-temporal Interferometric Synthetic Aperture Radar)技术对2018-2020年瀑布沟水电站的哨兵一号(Sentinel-1)图像进行时间序列形变监测，然后基于时间序列InSAR形变数据采用LSTM神经网络建立了形变预测模型，最终获取瀑布沟水电站的形变速率结果和时序形变的预测结果。结果表明，瀑布沟水电站最大沉降速率达到-34 mm/a-1，LSTM预测模型训练和测试过程中点尺度的均方根误差(RMSE Root Mean Squared Error)和绝对误差平均值(MAE, Mean Absolute Error)最小值分别为2.343 mm和2.010 mm，2.094 mm和1.654 mm。LSTM形变预测模型的预测结果显示2020年5月-9月的累计沉降值将达到71.29 mm。本研究证明了LSTM神经网络是一种有效InSAR时序形变预测方法。同时该模型的预测结果也可用于瀑布沟水电站的形变预警和辅助决策。
Deformation monitoring and prediction are crucial for early warning and timely remedial measures for abnormal situations in hydroelectric power stations. In this study, a Long Short-Term Memory (LSTM) neural network method was proposed to predict the time series deformation of the Pubugou Waterfall Power Station in the Dadu River Basin using Interferometric Synthetic Aperture Radar (InSAR) data. First, the Multi-temporal Interferometric Synthetic Aperture Radar (MT-InSAR) technique was used to monitor the time series deformation of the Pubugou Waterfall Power Station based on Sentinel-1 images from 2018 to 2020. Then, an LSTM neural network was employed to establish a deformation prediction model based on the time series InSAR deformation data. The results show that the maximum subsidence rate of the Pubugou Waterfall Power Station reaches -34mm/a-1. The point-scale RMSE and MAE of the LSTM prediction model during training and testing are the minimum values of 2.343mm and 2.010mm, and 2.094mm and 1.654mm, respectively. The predicted results of the LSTM deformation prediction model indicate that the cumulative subsidence from May to September 2020 will reach 71.29mm. This study demonstrates that the LSTM neural network is an effective method for predicting InSAR time series deformation. Moreover, the predicted results of this model can also be used for deformation warning and decision-making support for the Pubugou Waterfall Power Station.
黄会宝,江德军,刘恒,等. 基于多时相InSAR技术的大渡河瀑布沟水电站形变监测与预测[J]. 科学技术与工程, 2023, 23(30): 13112-13120.
Huang Huibao, Jiang Dejun, Liu Heng, et al. Deformation monitoring for Pubugou hydropower station of Dadu River base on multi-temporal InSAR technology[J]. Science Technology and Engineering,2023,23(30):13112-13120.