基于EEMD-GA-BP模型的大坝变形监测数据预测

针对大坝自动监测数据序列存在的不稳定性和测值漂移问题,提出了基于集合经验模态分解(EEMD)和遗传(GA)BP神经网络的大坝变形监测数据预测方法。采用EEMD技术提取反映大坝真实变形的低频信号,剔除自动监测系统数据中存在的噪声和野值,利用遗传算法优化的BP神经网络对真实信号进行学习与外推,据此构建EEMD-GA-BP模型。利用本文模型计算得到大坝变形的预测值,将其与实测变形值进行对比,并根据残差大小比较了本文方法与其它方法的预测效果。算例表明,本文提出的组合模型能有效地提高大坝变形预测精度。

Prediction of Dam Deformation Monitoring Data Based on EEMD-GA-BP Model

A prediction model of dam deformation monitoring data integrating Ensemble Empirical Mode Decomposition (EEMD), Genetic Algorithm (GA) and Back Propagation (BP) neural network is built to tackle the unstable performance and the drift of measured value of automatic monitoring data of dam deformation. The EEMD is used to extract the low-frequency signals which reflect the true deformation of dam and to remove the noise and outliers in the data of the automatic monitoring system; the GA-optimized BP neural network is employed to learn and extrapolate the real signals. The model-predicted deformation values are compared with measured values and also predicted values of some other methods in terms of residual error. Case study demonstrates that the proposed model could improve the prediction accuracy of dam deformation effectively.