...
  • 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Bulletin of engineering geology and the environment >Landslide displacement prediction based on a novel hybrid model and convolutional neural network considering time‑varying factors
【24h】

Landslide displacement prediction based on a novel hybrid model and convolutional neural network considering time‑varying factors

机译：考虑时变因子的新型混合模型和卷积神经网络的滑坡位移预测

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Accurate landslide displacement prediction is essential for an early warning system. At present, the inputs of the data-driven models adopted in landslide displacement prediction remain unchanged. However, considering that the sensitivities of landslide displacement states to external factors are constantly changing, it is reasonable to take trigger factors' time-varying characteristics into account for obtaining better prediction results. In this study, the input cumulative displacement was first decomposed by singular spectrum analysis, and the k-means algorithm was developed to cluster the periods representing the different states of the examined landslide. Then, a one-dimensional convolutional neural network with a detailed structure was introduced to consider time-varying inputs for periodic displacement prediction. A grey power model optimized by particle swarm optimization was proposed to predict trend displacements with less empirical judgement. Model's performance was mainly validated based on the Baishuihe landslide in the Three Gorges Reservoir area. The application results demonstrate that (i) the periods representing different landslide states can be obtained reasonably by clustering trend information; (ii) the optimized grey power model can be utilized more universally than the polynomials for trend displacement prediction; and (iii) the consideration of time-varying trigger factors in the data-driven model further enhances the model's prediction accuracy and robustness.
机译：精确的滑坡位移预测对于预警系统至关重要。目前，滑坡位移预测中采用的数据驱动模型的输入保持不变。然而，考虑到山体滑坡位移状态对外部因素的敏感性不断变化，因此考虑到获得更好的预测结果，采取触发因素的时变特征是合理的。在该研究中，输入累积位移首先通过奇异频谱分析分解，并且开发了K-Means算法以聚集代表被检查的滑坡的不同状态的时间。然后，引入了具有详细结构的一维卷积神经网络，以考虑用于周期性位移预测的时变输入。提出了一种通过粒子群优化优化的灰色电力模型，以预测具有较少实证判断的趋势位移。模型的性能主要基于三峡库区的白石河滑坡主要验证。申请结果表明，（i）可以通过聚类趋势信息合理地获得代表不同滑坡状态的时段; （ii）优化的灰色电力模型可以比趋势位移预测的多项式更普遍地利用; （iii）考虑数据驱动模型中的时变触发因子进一步提高了模型的预测准确性和鲁棒性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号