基于回归-ELM神经网络模型的滑坡变形及失稳预测模型

为准确预测滑坡的变形趋势,有效预防滑坡灾害的发生,提出了基于变形预测和检验的趋势判断模型。首先,利用回归分析,拟合得到滑坡的变形曲线,再利用组合权值,实现拟合结果的组合,得到滑坡变形的初步预测结果;其次,利用极限学习机(ELM神经网络)对初步预测结果进行误差修正,将修正结果与初步预测结果进行叠加,得到滑坡变形的综合预测值;最后,利用秩相关系数检验与Mann-Kendall检验,对滑坡变形趋势进行判断,以验证预测结果的准确性。经过实例检验得出,预测模型的预测效果较好,其组合预测及误差修正均能不同程度地提高预测精度及稳定性,且两检验模型的结果均与预测结果相符,相互验证了其可靠性。因此,预测模型能对滑坡变形趋势进行综合判断,为滑坡的变形研究提供了一种新的思路。

Prediction model of landslide deformation and instability based on regression-ELM neural network model

In order to predict the deformation trend of landslide accurately and prevent the occurrence of landslide effectively, a trend judgment model based on deformation prediction and test is put forward. Firstly, regression analysis is used to fit the deformation curve of the landslide, and combined weights is used to achieve the combination of the fitting results, obtaining the preliminary results of landslide deformation prediction; secondly, extreme learning machine (ELM neural network) is used to correct the error of the initial forecast results, then the corrected results and the preliminary prediction results are processed together, so that the comprehensive prediction value of the landslide deformation is obtained; finally, the rank correlation coefficient test and Mann-Kendall test are used to estimate the trend of landslide deformation to verify the accuracy of the prediction result. The test shows that the prediction model is good, the combination forecasting and error correction both can improve prediction accuracy and stability in some degree, and the two model test results and the prediction results are consistent, which verifies each other''s reliability. The prediction model can comprehensively judge the trend of landslide deformation, which provides a new way for the study of landslide deformation.