基于FEEMD-GRU-FC模型的滑坡位移预测

Landslide displacement prediction based on FEEMD-GRU-FC model

水库库区滑坡同时受到降雨和水库水位周期性变化的影响,这会干扰滑坡位移预测模型在该类区域的预测性能。为此,提出了一种针对环境周期性变化特点的FEEMD-GRU-FC(Fast Ensemble Empirical Mode Decomposition-Gated Recurrent Unit-Fully Connected Layer)滑坡位移预测模型。该模型充分考虑位移数据不同频率之间的噪声影响,将位移分解为不同频率的子序列分别预测,采用PCC(Pearson Correlation Coefficient)方法选择影响较大的外部环境因素,添加全连接层完善GRU模型提取环境数据中周期性变化特征的能力,模型结合三峡库区八字门滑坡的实际数据进行模拟仿真,使用不同预测模型和评价指标进行多角度对比验证。实验结果表明:所提模型在八字门滑坡的预测效果要优于其他3种模型,其拟合度达到了98.6%,3种不同误差指标均为最低,证明了该模型可以准确预测库区滑坡位移。研究成果可为防灾减灾工作提供指导。

Landslides in reservoir areas are affected by rainfall and periodic changes in reservoir water level,which will interfere with the prediction performance of landslide displacement prediction models.A FEEMD-GRU-FC(fast ensemble empirical mode decomposition-gated recurrent unit-fully connected layer)landslide displacement prediction model based on the periodic environmental changes was proposed.The model fully considered the influence of noise among different frequencies of displacement data,divided the displacement into sub-sequences of different frequencies to predict them respectively,and used the PCC(Pearson correlation coefficient)method to select external environmental factors with large influence.The ability of the GRU model to extract periodic change features from environmental data was improved by adding a full connection layer.The model was applied by combining with the actual data of the Bazimen Landslide in the Three Gorges Reservoir area.Different prediction models and evaluation indicators were used for multi-perspective comparison and verification.The experimental results showed that the model proposed was better than the other three models in predicting Bazimen Landslide displacement.Its fitting degree was 98.6%,and three kinds of error indexes were the lowest,which proved that the model can accurately predict the landslide displacement in the reservoir area.The research results can guide disaster prevention and reduction.