融合多测点数据相关性的大坝监测历史数据填补

Dam monitoring historical data filling algorithm based on fusion of multi-measurement point data correlation

基于历史数据的大坝安全监测预警、预报算法对数据集的质量要求较高,含有缺失值的数据集会明显降低算法结果的准确性。为提升数据质量,提出了一种融合多测点数据相关性的缺失值填补算法:基于各测点时间尺度的相关性,将满足一定相关度的时间序列作为预测模型的输入项,并引入迭代技术实现多测点的缺失值自动填补。为验证该算法对实际工程中不同类型缺失数据集的适用性,依据实测数据样本在缺失率、集中度、离散程度3个层次上共构造了12种不同类型的缺失数据集并进行试验。结果表明:针对不同类型的缺失数据集,该算法的RMSE均值在填补精度上较传统填补算法提升15%以上,nMAPE均值提升1%以上。

Algorithms for early warning and forecasting of dam safety monitoring based on historical data has high requirements on the quality of datasets,and the datasets containing missing values will significantly reduce the accuracy of the algorithm results.In order to improve the quality of data,this paper proposed a missing value filling algorithm that integrates the multi-measurement point data correlation.Based on the correlation of the time scale of each measuring point,the time series that satisfies a certain degree of correlation was used as the input of the prediction model,and the iterative technique was introduced to realize the automatic filling of the missing values of multiple measuring points.In order to verify the applicability of the algorithm to different types of missing datasets in practical engineering,according to the measured data samples,a total of 12 different types of missing datasets were constructed at three levels of missing rate,the concentration degree,dispersion degree and experiments were carried out.The results showed that for different types of missing datasets,the RMSE mean value of the algorithm was more than 15%higher than that of the traditional filling algorithm,and the nMAPE mean value was more than 1%higher.