徐丛, 王少伟, 顾冲时, 苏怀智. 融合空间关联性的特高拱坝位移概率性预测模型[J]. 武汉大学学报 ( 信息科学版), 2023, 48(3): 433-442. DOI: 10.13203/j.whugis20200508
引用本文: 徐丛, 王少伟, 顾冲时, 苏怀智. 融合空间关联性的特高拱坝位移概率性预测模型[J]. 武汉大学学报 ( 信息科学版), 2023, 48(3): 433-442. DOI: 10.13203/j.whugis20200508
XU Cong, WANG Shaowei, GU Chongshi, SU Huaizhi. A Probabilistic Prediction Model for Displacement of Super High Arch Dams Considering Deformation Spatial Association[J]. Geomatics and Information Science of Wuhan University, 2023, 48(3): 433-442. DOI: 10.13203/j.whugis20200508
Citation: XU Cong, WANG Shaowei, GU Chongshi, SU Huaizhi. A Probabilistic Prediction Model for Displacement of Super High Arch Dams Considering Deformation Spatial Association[J]. Geomatics and Information Science of Wuhan University, 2023, 48(3): 433-442. DOI: 10.13203/j.whugis20200508

融合空间关联性的特高拱坝位移概率性预测模型

A Probabilistic Prediction Model for Displacement of Super High Arch Dams Considering Deformation Spatial Association

  • 摘要: 基于机器学习语言建模时,传统方法仅以拟合均方误差(mean square error,MSE)最小为单一优化目标,容易引起过拟合问题。对此,基于关联向量机(relevance vector machine,RVM),建立了融合变形空间关联性的双优化目标约束下的概率性预测模型。利用形状相似度(shape similarity index,SSI)对拱坝变形的空间关联性进行量化,并将单测点MSE和区域变形SSI相融合,共同作为RVM模型的训练优化目标,以期实现MSE尽可能小,而SSI尽可能大。以锦屏一级拱坝为例,预测均方根误差和最大误差的平均降幅分别为31.2%和24.8%,使用多核函数之后,模型预测性能进一步提升;RVM模型的预测置信带宽明显小于多元回归模型,平均降幅为75.1%,这表明双目标RVM模型可有效提升特高拱坝位移预测的精度和稳定性,并降低不确定性。

     

    Abstract:
      Objectives  Machine learning language has become an ideal modeling tool in the field of dam health monitoring with its powerful nonlinear data mining ability. However, the minimum fitting mean square error (MSE) is determined as the only optimization objective in the traditional modeling process, which is likely to cause over-fitting problems.
      Methods  To overcome this problem, based on the relevance vector machine (RVM), a probabilistic prediction model is established under the constraint of double optimization objectives, which integrates the deformation spatial association and MSE. The deformation spatial association is quantified by the shape similarity index (SSI) at first. The double objective is then established with the combination of the MSE and SSI, and is achieved by making the MSE as small as possible, while the SSI is as large as possible.
      Results  Engineering example of the Jinping-I arch dam shows that the average decrease proportion of the root mean square error and maximum absolute error of the proposed double objective RVM model is 31.2% and 24.8%, respectively, and the prediction performance can be further improved by using the multi-kernel function. The prediction confidence bandwidth of the RVM model is significantly smaller than that of the traditional multiple linear regression model, with an average decrease proportion of 75.1%.
      Conclusions  Therefore, the multi-kernel double objective RVM model established for the displacement of super high arch dams can effectively improve the prediction performance and reduce the uncertainty.

     

/

返回文章
返回