Study On Dam Safety Monitoring Statistic Models Based On Partial Least Squares Regression

According to previous dam safety monitoring statistic models and focus on the defects of least square regression statistic theory, the statistic models for dam safety monitoring were put forward in the dissertation based on the partial least square square regression, orthogonal signal correction, weighted block recursive partial least squares regression, neural network. The performances of the models were analyzed in detail through the Heihe Jinpen reservoir measured data. The study was not only of practical importance for hydraulic projects, but also of vital significance for promoting dam safety control level in our country.The contents and major results are as follows:(1) The previous dam safety monitoring statistic models were comprehensively and systematically summarized, and Through the engineering practice the problems on least squares regression in formulating statistic model structure were analyzed, based on which it was pointed out that the multi-relativity among factors of least squares regression model was the key matter in causing structure instability and unclear explanation.(2) Based on partial least square regression theory and aiming at statistic model formulating, the statistic model of partial least square regression was presented. It was indicated through case study that the model could effectively overcome the serious multi-relativity among factors; consequently it could make the structure of statistic model stable and enhance the result explanation. It was shown by comparing of modeling results and project data that the model was a powerful tool for formulating basic structure under the condition of multi-relativity among factors. It was shown by comparing of modeling results and project data that the model was a powerful tool for formulating basic structure under the condition of multi-relativity among factors. Therefore, the innovative results of the study were stable model structure formulated on the basis of partial least square method and ageing figure as a important method in evaluation of dam safety.(3) Firstly, orthogonal signal correction was applied to pretreat independent, then using partial least squares regression through the engineering practice to formulate partial least squares regression statistic model based on orthogonal signal correction. The study indicatedbecause the orthogonal factors of independent were eliminated by orthogonal signal correction, only one principal component was needed in using partial least squares to formulate model which was more simple on structure and had better explanation than ordinary least squares.(4) The ordinary partial least squares had following shortcomings: If the model was formulated, it would be fixed. When the process characteristics or operating conditions were changed, the modle could be updated timely. Aiming at the shortcoming, the dam safety monitoring statistic model was fonnulated firstly by weighted block recursive partial least squares regression. The practical study indicated the model could be well adapted to the changes in dam operations, and make a more reasonable explanation for dam safety state.(5) In view of the complex nonlinear of dam safety monitoring system, In this dissertation, RBF neural network was combined to partial least squares regression, and corresponding dam safety monitoring statistic model was formulated firstly which more accurately described relationship between dam factors and independent variable.