学术文献库

In the framework of risk assessment in nuclear accident analysis, best-estimatecomputer codes, associated to a probabilistic modeling of the uncertain input variables,are used to estimate safety margins. A first step in such uncertainty quantificationstudies is often to identify the critical configurations (or penalizing, in thesense of a prescribed safety margin) of several input parameters (called ``scenarioinputs''), under the uncertainty on the other input parameters. However, the largeCPU-time cost of most of the computer codes used in nuclear engineering, as theones related to thermal-hydraulic accident scenario simulations, involve to develophighly efficient strategies. This work focuses on machine learning algorithms bythe way of the metamodel-based approach (i.e., a mathematical model which is fittedon a small-size sample of simulations). To achieve it with a very large numberof inputs, a specific and original methodology, called ICSCREAM (Identificationof penalizing Configurations using SCREening And Metamodel), is proposed. Thescreening of influential inputs is based on an advanced global sensitivity analysistool (HSIC importance measures). A Gaussian process metamodel is then sequentiallybuilt and used to estimate, within a Bayesian framework, the conditionalprobabilities of exceeding a high-level threshold, according to the scenario inputs.The efficiency of this methodology is illustrated on two high-dimensional (arounda hundred inputs) thermal-hydraulic industrial cases simulating an accident of primarycoolant loss in a pressurized water reactor. For both use cases, the studyfocuses on the peak cladding temperature (PCT) and critical configurations aredefined by exceeding the 90%-quantile of PCT. In both cases, the ICSCREAMmethodology allows to estimate, by using only around one thousand of code simulations,the impact of the scenario inputs and their critical areas of values.