Research On Bayesian Hydrologic Uncertainty Processor And Its System Integration Method

Floods and droughts are the most widespread natural disasters in the world with the most significant harm to human survival and development.Understanding the objective law of hydrological process is helpful to the rational and efficient utilization of water resources,flood control,drought relief and disaster reduction under global climate change.However,existing deterministic hydrological models cannot accurately describe hydrological processes with complex internal connections.Most of them have uncertainties in many aspects,such as input data,model parameters,and model structure.The single model results obtained from these deterministic hydrological models are difficult to fully describe the evolutionary characteristics of hydrological processes.Therefore,there is an urgent need to study the quantitative methods of hydrological uncertainty,which are applied easily in engineering,analyzing the statistical rules of uncertainty,realizing the quantitative analysis of water resources management decision-making risk and providing important decisionmaking basis for flood risk analysis,flood control,drought relief and disaster mitigation.Bayesian Hydrological Uncertainty Processor(HUP)has been widely used in the quantification research of hydrological uncertainty.This paper focuses on some key problems in the theoretical and engineering applications of HUP,probabilistic curve fitting,Bayesian conjugate distribution,normal quantile conversion,statistical hypothesis testing and other theories and methods which have been applied in HUP.Simultaneously,some meaningful theoretical results have been achieved in simplifying the numerical solution process of HUP and perfecting the theoretical derivation in this paper.Meanwhile,to apply various professional models of water conservancy including HUP to the engineering practice of river basin management,research on the integration method of water resources management system is carried out,and a set of water resources management system integration framework is established.The main contents and innovations of the paper are as follows:(1)In view of the problems of low accuracy and strong subjectivity in the marginal distribution fitting of hydrological data,the model principle of HUP has been analyzed,the Gaussian Mixture Model(GMM)with arbitrary precision fitting capability is introduced as the marginal distribution fitting of hydrological data structure.And then,this paper obtains a processor that unifies the marginal distribution estimation model and solution method of the measured and forecasted data,which is called the hydrological uncertainty processorGaussian mixture model(HUP-Gaussian Mixture Model,HUP-GMM).The research on the prediction results of the Xin'anjiang hydrological model at Yichang station shows that compared with the HUP model that uses the conventional probability distribution function to estimate the edge distribution,the peak value of the HUP-GMM posterior distribution density curve is higher,the overall shape is more obviously concentrated in the peak neighborhood,and the quantification performance of hydrological uncertainty of HUPGMM is better.(2)For the problem of complexity of the Bayesian formula derivation and solution in HUP,which is not conducive to the theoretical expansion and practical application of HUP,the research work explores the application of Bayesian conjugate distribution theory in HUP and constructs a new correlation between variables.Based on HUP-GMM,a model that uses the normal function structure to directly fit the posterior distribution(HUP-Linear Gaussian Mixture Model,HUP-LG)is derived.The research results show that HUP-LG and HUP-GMM have the same ability to quantify the hydrological uncertainty,but the theoretical derivation and solution process of HUP-LG has been greatly simplified.(3)For variables derived from the Normal Quantile Transform(NQT)in HUP,the hypothesis that the residual term of the correlation formula follows the normal distribution lacks theoretical basis.Therefore,based on HUP-LG,a secondary normal quantile transformation of the residual term is derived.And then,the actual distribution is solved according to the secondary NQT space variable.Finally,the improved model of the conditional distribution of the original spatial measured data(HUP-Double Linear Gaussian Mixture Model,HUP-DLG)is solved.The research results show that HUP-DLG improves the theoretical completeness of the HUP model,and further improves the hydrological uncertainty quantification performance of HUP.(4)Starting from the high-performance system integration of various professional models of water conservancy,including HUP,HUP-GMM,HUP-LG and HUP-DLG,the traditional water resources model organization method and the centralized system integration framework have poor scalability,the system performance restricted by the operating efficiency of a single model,and the difficulty in achieving high efficiency and stable integration of the system.The research work proposes a method for constructing a water resources management model library that can effectively manage and efficiently integrate professional models.And then,an efficient clustering method of water resources management system services is established.The actual engineering application results show that the proposed method significantly improves the operation efficiency and system interaction performance of professional model services of water resources management,and has the advantages of sustainable integration,cross-platform,and rapid response.