摘要
大坝变形的时空演变预测分析有助于大坝管理人员及时掌握大坝空间的整体变形状态。目前,大坝变形预测研究分为两个方面:1)通过仅对分布变形仪器部位进行时间序列预测,得出下一时刻的变形值(如BP神经网络);2)利用周围变形数据进行空间插值,得到当前时刻未分布仪器点的变形值。单独使用上述任何一种方法都无法利用历史变形数据预测下一时刻未分布仪器部位的变形状况。针对该问题,结合空间预测模型时空克里金方法(STKriging,STK)与神经网络模型即BP神经网络及门限循环神经网络(Gated Recurrent Unit,GRU)各自的优势,构造了一种新型时空序列预测算法(BP-STK-GRU),实现了对未分布监测仪器部位的变形值预测。主要步骤包括:1)GRU优化单个测点的历史时间序列变形值;2)BP拟合测点下一时刻数据的整体趋势;3)利用STK拟合BP预测结果的稳定部分;4)结合空间插值及BP空间整体预测值,得出未分布仪器点的变形值。实验结果表明,所提方法是有效的,并且在对未知点的变形预测稳定性及精确度方面都有很好的表现。
The analysis of the spatial-temporal evolution of dam deformation is conducive for managers to master the overall deformation of the dam’s space.The existing predictive research on dam deformation can be divided into two parts.The first part is only making time series prediction for instrument part with distribution deformation,and the se-cond part is using a method of spatial interpolation at the current moment to obtain unknown point’s value of deformation.Both of these cannot use the historical deformation time series data to predict the deformation of the undistributed instrument.To solve this problem,combining the advantages of traditional spatial-temporal prediction model(STK)and neural network modelssuch such as BP and Gated Recurrent Unit(GRU),this paper constructed a spatio-temporal sequence prediction algorithm named BP-STK-GRU.The main steps are described as follows.Firstly,GRU optimizes the historical time series of individual measuring points.Secondly,BP fits the overall trend of spatio-temporal data at measuring points of the next moment.Thirdly,STK fits the stable parts of BP prediction results.Lastly,the spatial residual value and the overall BP space prediction are combined to get the deformation of the undistributed instrument.The experimental results show that the method is effective,and it has good performance in predicting the stability and accuracy of the deformation value of the unknown point.
作者
毛莺池
曹海
何进锋
MAO Ying-chi;CAO Hai;HE Jin-feng(College of Computer and Information,Hohai University,Nanjing 211100,China)
出处
《计算机科学》
CSCD
北大核心
2019年第2期223-229,共7页
Computer Science
基金
"十三五"国家重点研发计划项目(2018YFC0407105)
华能集团重点研发课题(HNKJ17-21)资助
关键词
大坝变形
时空预测
门限循环神经网络
BP网络
空间插值
Dam deformation
Spatio-Temporal prediction
Gated recurrent unit
BP network
Spatial interpolation
