Study On Missing Value Processing Method Of Gravity Dam Deformation Monitoring Data

With the rapid development of water conservancy project in China,the task of dam safety monitoring is also increasing.As a comprehensive variable reflecting the safety state of gravity dam,the deformation value is an important indicator to evaluate the structural performance.The monitoring and analysis of the deformation value of the dam body can better understand the deformation law of the dam body and predict the deformation trend of the dam body,which is of great significance to the safe and stable operation of the gravity dam.However,the dam operation faces many uncertain factors and the monitoring period is long.In the process of data acquisition,data transmission and data storage,such as instrument damage and network interruption will lead to data loss.Because the current data analysis methods and procedures are based on the completeness of data,the reliability of the analysis results will be reduced when there is a lack of data,which is not conducive to the later dam safety evaluation.Therefore,how to effectively interpolate the missing values in deformation monitoring data to obtain a higher quality data set is of great significance for improving the reliability of dam safety monitoring and early warning prediction results.In this paper,for different types of missing values,based on the spatial and temporal distribution law of gravity dam deformation,the data repair models are constructed by using the long short-term memory neural network(LSTM)of the deep learning model to interpolate the missing values.The main research contents and results are as follows :(1)Aiming at the problem of missing value in the deformation monitoring data of gravity dam,through the analysis of a large number of actual monitoring data,the causes,modes,generation mechanism and traditional missing value processing methods are studied.At the same time,the calculation of engineering examples shows that when there is a missing value in the training set of the gravity dam monitoring model,the accuracy of the prediction model will decrease significantly,which cannot accurately reflect the real displacement of the measuring point.It proves the necessity of interpolation of the missing value of the monitoring data.(2)Aiming at the problem of single measuring point data missing,the correlation analysis was carried out based on the maximum information coefficient(MIC),and the relationship between the same type of effect quantity and the effect quantity and the environment quantity in the deformation monitoring data was quantified.The long short-term memory neural network(LSTM)was introduced to mine the characteristic information between different variables,and the repair model of deformation monitoring data missing value based on MIC-LSTM was established.The results show that the repair accuracy of missing data in the model is significantly improved after the input factors are optimized by MIC correlation analysis.When selecting the input factors of the model,the data of the same type of measuring points should be selected as independent variables for analysis,and the measuring points with the highest correlation coefficient should be selected as the input factors of the model according to the correlation analysis results.(3)Aiming at the problem of multi-point data missing,decomposition of deformation sequence using adaptive noise ensemble empirical mode decomposition(CEEMDAN),and the complexity of different scale components is evaluated based on permutation entropy(PE)algorithm.In order to reduce the calculation scale and error,the components with similar complexity are reorganized,and the multi-point data missing value repair model of gravity dam based on CEEMDAN-PE-LSTM is established.The calculation results show that compared with the traditional prediction model,the multi-point interpolation model proposed in this paper has higher repair accuracy for the missing value of deformation monitoring data,and can accurately reflect the actual change law of the monitoring value of the missing data segment.The model has good generalization and universality.(4)In order to verify the effectiveness of missing value interpolation to improve the accuracy of the prediction model,four different missing data sets are constructed.The missing data sets are interpolated by the M-L and C-P-L missing value interpolation methods proposed in this paper and the traditional missing value processing methods.Based on the data sets obtained by different interpolation methods,the traditional prediction model support vector machine(SVM),BP neural network and LSTM are used to predict.The results show that the data set obtained by M-L and C-P-L missing value interpolation methods have high accuracy and small error in prediction,and the stability is good at different missing rates,indicating that these two methods can be effectively used to improve the quality of monitoring data.The effectiveness and universality of the interpolation method proposed in this paper are further proved by comparing and analyzing the results of multiple measuring points.