Research On Prediction Of The Return Value Of Marine Extreme Environment

The natural environment in which ships and marine structures are located is extremely complex and harsh,and they are often subjected to various loads from the marine environment,such as waves,wind speed,tides,etc.It is necessary to consider the extreme values of the multi-year return period of the relevant marine environment during design to cope with the possible extreme loads.Therefore,based on the actual historical observation data,it is important to establish the marine extreme environment prediction method and realize the prediction of extreme values of marine environment elements such as wind speed,wave height and tide level values to ensure the safety and reliability of ships and marine structures during their service life.Due to the complexity and uncertainty of the marine environment,the data from a single observation point are affected by various factors leading to errors in the estimation of extreme reproduction values,while the marine data from multiple nearby observation points can often provide more available information about a region or a sea area,which helps to improve the accuracy and reliability of the extreme value prediction,so it is crucial to consider multi-site marine environmental data in extreme value prediction.In this thesis,we propose a method to test the reproducible values of marine environment elements under specific conditions by taking the tide level data of Kansai region in Japan as an example and using the joint analysis of the polar dependence and spatial and temporal characteristics of multi-site observation data,and establish a more accurate method to predict and test the reproducible values of marine environment at multi-site,and compare and verify them by example.Meanwhile,based on the Brest tide level data in France,we explored various ways of univariate extreme value modeling,and demonstrated the feasibility of the mixed distribution method of extreme values for extreme value modeling of marine environmental data,the main research contents of this thesis are as follows:(1)Firstly,univariate extreme value analysis is performed on the measured tide level data,and four univariate extreme value modeling approaches are considered on the basis of pre-sampling the data in groups,and the comparison reveals that the application of GEV_rdistribution under the order statistic in the extreme value analysis of marine environmental elements with long time span helps to reduce the confidence interval of extreme value prediction.Meanwhile,different parameter estimation methods are compared,and the applicability of the method of great likelihood estimation(MLE)for parameter estimation of extreme value distribution is demonstrated.(2)To address the key problem that the order value r of the GEV_r distribution under the order statistic method cannot be reasonably selected,the commonly used multiplier test and entropy difference test methods are introduced,and a relevant program based on R language is written to propose a graphical assisted selection method of r value based on the change of the standard deviation of the parameter estimates,which is simpler to realize the selection of r value.(3)Aiming at the inaccurate threshold selection of POT method commonly used in the prediction of return value,the mixed distribution theory of extreme values was applied to the extreme value modeling of Marine environmental data.According to the characteristics of the mixed distribution of extreme values,a new distribution is formed by combining the specific subject distribution with the tail of GP distribution.The appropriate threshold is selected through the fitting effect of the overall data,so as to depict the whole process of Marine environmental data more accurately and improve the accuracy of POT method’s extreme value analysis.Taking the tidal level of different periods in Brest region of France as an example,it is found that the occurrence frequency of extreme tidal level events in this region has increased significantly in recent years.The prediction result of the return value obtained by the extreme mixed distribution method is related to the main data distribution and data volume.(4)Aiming at the coupling problem of multi-site Marine data,the multivariate conditional extreme value solution is applied to the analysis of multi-site environmental data.The mixed extreme value distribution method is used to select the threshold of the marginal distribution of single site,and the Laplacian marginal transformation is carried out to establish the conditional extreme value model of multi-site tide data.Taking the multi-station tidal level along the coast of Kansai region of Japan as an example,a test method for the reproducible value under multi-station extreme conditions is presented based on the conditional extreme value model and the dependence of multi-station extreme value through R language program combined with Monte Carlo simulation.