Monitoring Minimum DNBR Using a Support Vector Regression Model

Abstract

The pressurized water reactor operates in the nucleate boiling regime. The transition from nucleate boiling to the film boiling accompanied by severe reduction of the heat transfer capability can result, however, in a boiling crisis that in the long run can cause fuel cladding melting. Therefore, it is very important to predict and monitor the departure from nucleate boiling (DNB) to prevent fuel clad melting and control the boiling crisis. In this study, the minimum DNB ratio (MDNBR) is predicted based on support vector regression (SVR) model using a number of measured signals from the reactor coolant system. SVR models are trained using a training data set and verified against test data set, which does not include training data. The SVR models have been applied to the first cycle of the Yonggwang 3 nuclear power plant. The estimation accuracy of the MDNBR was high enough to be used in DNB monitoring. Also, SVR model provides larger MDNBR values as compared to the existing core operation limit supervisory system, which allows greater operation margin.