基于TW-SVM预测模型的某堆石坝变形预测分析

TW-SVM prediction model-based analysis on deformation of a rockfill dam

为提高监测资料有缺失的大坝变形预测模型精度,采用支持向量机方法建立一种具有小样本、高维、非线性的预测模型,并结合对其重要组成部分核函数的分析应用,提出一种根据结构风险最小化的TW-SVM预测模型。以某堆石坝为例进行研究,利用坝坡垂直位移和水平位移的监测数据,分别采用TW-SVM方法和BP神经网络(NET)方法建立相应预测模型进行比较分析。结果表明:采用TW-SVM方法和NET方法预测的垂直位移最大绝对误差分别为0.58 mm和6.18 mm,最大相对误差分别为270.00%和1 286.22%;采用TW-SVM方法和NET方法预测的水平位移最大绝对误差分别为0.25 mm和14.91 mm,最大相对误差分别为31.25%和1 189.85%;TW-SVM预测模型比NET预测模型更适合于影响因素为时间、水位的小样本预测分析。研究结果为堆石坝变形预测与分析提供参考。

In order to improve the accuracy of the deformation prediction model for the monitoring data absent dam,a prediction model with the merits of small samples,high dimension and nonlinearity is established with the method of support vector machine（SVM）,and then a TW-SVM prediction model based on structural risk minimization is proposed. By taking a rockfill dam as the study case,the relevant prediction models established with both the TW-SVM Method and the BP neural network method respectively for the comparative analysis concerned. The result shows that the maximum absolute errors of the vertical displacements got from both the TW-SVM Method and the BP neural network method are 0. 58 mm and 6. 18 mm respectively,while the maximum relative absolute errors are 270. 00% and 1 286. 22% respectively as well. Meanwhile,the maximum absolute errors of the horizontal displacements got from both the TW-SVM Method and the BP neural network method are 0. 25 mm and14. 91 mm respectively,while the maximum relative absolute errors are 31. 25% and 1 189. 85% respectively as well,thus if compared with the NET prediction model,the TW-SVM prediction model is more suitable for the small sample prediction analysis for the impacting factors of time and water level. Generally,the study result can provide a reference for the prediction and analysis made on deformation of rockfill dam.