A Combination Model for Displacement Interval Prediction of Concrete Dams Based on Residual Estimation
A Combination Model for Displacement Interval Prediction of Concrete Dams Based on Residual Estimation
复制标题
DOI：
10.3390/su142316025
复制
发表时间：
2022-11
期刊：
影响因子：
3.9
通讯作者：
Xin Yang;Yan Xiang;Guangze Shen;Mengyi Sun
中科院分区：
文献类型：
--
作者：
Xin Yang;Yan Xiang;Guangze Shen;Mengyi Sun
大坝位移的准确预报和合理预警是大坝安全监测的重要内容。然而，基于确定性点预测结果识别异常位移是困难的。针对这一问题，本文提出了一种集成多种策略的模型，以实现大坝位移的高精度点预测和区间预测。具体而言，大坝位移的区间预测分三个阶段实现。在第一阶段，基于极值梯度提升（XGBoost）的位移预测模型的构建。第二阶段，利用本文提出的残差估计方法生成XGBoost模型的预测误差序列，通过最大似然估计方法构建基于人工神经网络（ANN）的残差预测模型。第三阶段，对神经网络模型的训练误差组成的噪声序列进行区间估计。最后，将上述结果综合起来，实现大坝位移的区间预测。通过某混凝土坝的监测数据验证了模型的有效性。结果表明，混合模型不仅可以获得比单一模型更好的点预测精度，而且可以提供高质量的区间预测结果。
Accurate prediction and reasonable warning for dam displacement are important contents of dam safety monitoring. However, it is difficult to identify abnormal displacement based on deterministic point prediction results. In response, this paper proposes a model that integrates several strategies to achieve high-precision point prediction and interval prediction of dam displacement. Specifically, the interval prediction of dam displacement is realized in three stages. In the first stage, a displacement prediction model based on Extreme gradient boosting (XGBoost) is constructed. In the second stage, the prediction error sequence of XGBoost model is generated by the residual estimation method proposed in this paper, and the residual prediction model based on artificial neural network (ANN) is constructed through the maximum likelihood estimation method. In the third stage, the interval estimation of the noise sequence composed of the training error of the ANN model is carried out. Finally, the results obtained above are combined to realize the interval prediction of the dam displacement. The performance of the proposed model is verified by the monitoring data of an actual concrete dam. The results show that the hybrid model can not only achieve better point prediction accuracy than the single model, but also provide high quality interval prediction results.