Study On Statistics Models Of Gravity Dam Deformation Monitoring

With the construction of several giant dams, hydropower is playing an increasinglyimportant role in China’s energy structure. Meanwhile, the work of the dam behavior,diagnose the health of the dam has become a top priority dam staff. Dam their environmentmore complex, it may be some unknown adverse factors during operation, it is necessary fordam safety monitoring. Once the abnormal, timely analysis and find out why, you can gaintime to remedy the dam safety. Currently monitoring data collected in the dam has been moreadvanced, has been largely automated. However, modeling and analysis of the data sample isstill in the semi-theoretical, semi-empirical phase, for which monitoring data analysisbecomes more difficult. Deformation monitoring is the most important project of dam safetymonitoring, and it is safe or not be able to directly reflect the most reliable monitoring of theamount of the dam, while the statistical model because it is simple, intuitive and is used bythe majority of monitors. Therefore, this paper mainly statistical analysis of dam deformationmonitoring data made the following analysis:(1) This paper briefly introduced the dam deformation monitoring data collection methodused in the line of sight and forward intersection method. In reference to previous studiesbased on the analysis of the composition of the first model of dam deformation monitoringbasic independent variable factors, including hydraulic components, temperature and agingcomponents. Followed by a discussion of the current commonly used methods of statisticalanalysis-multiple regression analysis, the drawback is that multiple regression analysis did notconsider the problem of multicollinearity between the independent variables, for the multipleregression analysis to optimize get stepwise regression analysis, engineering examplesstepwise regression analysis to obtain satisfactory results.(2) Consideration stepwise regression analysis to improve the quality of dam safetymonitoring model is linearly related to abandon serious argument items, which theindependent variable factor for the displacement of the explanation is not particularlydesirable in the dam. Partial least squares method effectively avoids the disadvantages ofordinary multiple regression analysis and stepwise regression. Its basic principle is thehigh-dimensional data space projection arguments to the corresponding low-dimensionalfeature space, get the feature vector orthogonal to each other, and to emphasize theindependent variable on the dependent variable to explain and predict the feature vector isselected, and then build from a linear regression relationship between the feature vectors ofindependent and dependent variables between. Both to avoid collinearity problems, but alsoconsider the independent variable on the dependent variable explained. Finally, project examples demonstrate the partial least squares method to predict the stability of the damdeformation monitoring model.(3) BP neural network is a nonlinear dynamic system consists of a large number ofneurons in a certain topology generated during their training for the initial weights andthresholds randomly selected network is unstable or tends to fall into local minima. Geneticalgorithm selection, crossover and mutation are three ways of operating, maintaining stablemodel, and continue to generate new variation has a good global search capability. Thegenetic algorithm is applied to optimize the BP network initial weights and thresholds.Establishment of GA-BP model, validated by engineering examples, GA-BP model than thesimple BP neural network model has better accuracy and stability training.(4) Traditional dam safety monitoring model can only reflect a single measuring pointpartial dam deformation. This paper introduces the measuring point coordinates analysis ofmulti-dimensional model of dam monitoring measuring point multi-directional model,determined by partial least squares regression coefficient method to analyze the relationshipbetween the environment and the amount of deformation of the dam between. Examples ofmulti-dimensional analysis showed that the measuring point model has a better model thanthe one-sided point of generality and comprehensiveness.