基于M-ELM的大坝变形安全监控模型

针对大坝变形监测数据存在的非线性强、异常值诊断和剔除工作复杂及传统监控模型抗粗差能力差等问题,结合稳健估计理论抗粗差性强和极限学习机在处理非线性问题方面的优势,建立了基于稳健估计极限学习机的大坝变形安全监控模型。试验确定网络隐含层层数,构建4次方损失函数,采用加权最小二乘法计算输出权值,实现原始监测数据的拟合和预测。以某工程大坝变形监测数据为例进行建模分析,结果表明:以反映模型预测精度的均方误差和平均绝对百分误差及反映模型鲁棒性的中位数绝对偏差作为评价指标,基于稳健估计极限学习机的大坝变形安全监控模型的各项指标明显优于对比模型。

A safety monitoring model of dam deformation based on M-ELM

Aiming at the problems of strong nonlinearity, complexity of diagnosing and eliminating abnormal values, and poor ability to resist gross errors of traditional monitoring models for the analysis of dam deformation monitoring data, a dam deformation monitoring model based on robust estimate extreme learning machine(M-ELM)was established. It combines the theory of robust estimation which has strong roughness tolerance with the extreme learning machine which has strong ability in dealing with nonlinear problems. The network number of the hidden layers was determined by tests and a fourth power loss function was built. A weighted least square method was used to calculate the output weights, and the fitting and prediction of the original monitoring data was carried out. The monitoring data of a certain dam deformation was taken as an example for modeling analysis. The mean square error and the mean absolute percentage error reflecting the prediction accuracy, and the median absolute deviation representing the model robustness were taken as the evaluation indexes. The results show that each index of the dam deformation monitoring model based on robust estimate extreme learning machine is superior to other models.