基于多层次滤波降噪的IGA-NARX混凝土坝变形预测模型  

IGA-NARX Dam Deformation Prediction Model Based on Multi-level Filtering Noise Reduction

作　　者：杨孟[1] 李永福 梁云 陈艺征 顾冲时[1] YANG Meng;LI Yong-fu;LIANG Yun;CHEN Yi-zheng;GU Chong-shi(College of Water Conservancy and Hydropower Engineering,Hohai University,Nanjing 210098,China;State Grid Chongqing Electric Power Reserach Insitite,Chongqing 401123,China;State Grid Smart Grid Research Instite Co.,Ltd.,Beijing 102209,China)

机构地区：[1]河海大学水利水电学院,江苏南京210098 [2]国网重庆市电力公司电力科学研究院,重庆401123 [3]国网智能电网研究院有限公司,北京102209

出　　处：《水电能源科学》2024年第4期142-146,共5页Water Resources and Power

基　　金：国家电网公司总部科技项目(5108-202218280A-2-417-XG)。

摘　　要：针对混凝土坝变形数据中存在非高斯分布噪声污染,难以描述混凝土坝变形数据自身的趋势性、季节性的问题,采用CEEMD与粒子滤波法相结合的方法对锦屏一级大坝径向位移进行分析。先将CEEMD分解后的高频和低频分量进行区分,仅对高频分量进行粒子滤波降噪,再进行分量重构;通过多层次滤波降噪处理的位移数据驱动IGA-NARX神经网络构建预测模型,并使用R_(RMSE)、M_(MSE)等指标进行评价。工程实例验证表明,所提模型相较于对比模型在评价指标上均有一定提升,具有较好的实用价值。In view of the non-Gaussian noise pollution in the deformation data of concrete dam,it is difficult to describe the trend and seasonality of the deformation data of concrete dam.Combination of CEEMD and particle filter method is proposed to analyze the dam radial displacement of Jingpin Level I Hydropower Station.Firstly,the high frequency component and the low frequency component after CEEMD decomposition are distinguished,and only the high frequency component is denoised by particle filter.And then the component is reconstructed.The IGA-NARX neural network is driven by the displacement data of multi-level filtering and noise reduction,and the prediction model is constructed.The indexes of RRMSE and M MSE is used to evaluate the effectiveness of the model.It is verified by engineering examples.Compared with the comparison model,the proposed model has certain improvement in several evaluation indexes and has good practical value.

关 键 词：CEEMD分解 混凝土坝变形监测 粒子滤波降噪 NARX神经网络 IGA算法 

分 类 号：TV698.11[水利工程—水利水电工程]