| 摘要: |
| 水库大坝安全评价导则指出,监测资料应及时整编分析,保证监测数据完整性,以便通过监测资料及时了解大坝性状,并为大坝总体安全评价提供基本资料。传统的大坝缺失数据补全方法依靠于完整的前置数据和经验基函数,对于数据缺乏的中小型土石坝则效果不佳。利用经验模态分解算法分析缺失测点和同源测点数据,从较少的数据中提取有效信息。针对不同复杂度下分解得到的分量不统一问题,利用动态时间调整算法进行聚类整合。最后对聚类数据集分别建立基于门控循环单元的预测模型,构建了乏数据下历史监测数据EMD-GRU填补算法。基于实际工程监测数据对该算法和传统算法进行对比发现:均方误差降低至0.6以下,在乏数据的背景下比传统模型有更好的稳定性和泛化性。 |
| 关键词: 土石坝安全监测 缺失数据填补 乏数据 动态时间调整算法 经验模态分解算法 门控循环单元 |
| DOI： |
| 分类号: |
| 基金项目:国家重点研发计划资助项目（2022YFC3005502）；国家自然科学基金资助项目（52279135）；江西省水利厅重大科技项目（202124ZDKT06） |
|
| EMD-GRU-Based Missing Data Filling Method for Earth-Rock Dam Safety Monitoring Under Sparse Data Conditions |
|
ZhaoRuiQiao, LiDengHua, ShiBeiXiao
|
|
Nanjing Hydraulic Research Institute
|
| Abstract: |
| The guidelines for dam safety assessment in reservoirs state that monitoring data should be compiled and analysed in a timely manner to ensure the integrity of the monitoring data, so that the monitoring data can provide a timely understanding of the dam"s characteristics and provide basic information for the overall safety assessment of the dam. The traditional method of completing the missing data of dams relies on complete antecedent data and empirical basis functions, which is ineffective for small and medium-sized earth-rock dams with lack of data. The empirical modal decomposition algorithm is used to analyse the missing and homologous measurement point data to extract effective information from less data. For the problem of inconsistency of the components obtained from decomposition under different complexity, the dynamic time adjustment algorithm is used for cluster integration. Finally, the prediction model based on gated cyclic unit is established for the clustered dataset respectively, and the EMD-GRU filling algorithm is constructed for the historical monitoring data under the lack of data. Comparison between this algorithm and the traditional algorithm based on actual engineering monitoring data reveals that the mean square error is reduced to less than 0.6, and it has better stability and generalisation than the traditional model in the context of lack of data. |
| Key words: Earth-rock dam safety monitoring Missing data filling method Sparse data Dynamic Time Warping Empirical Mode Decomposition Gated Recurrent Unit |
