Optimization research of heat transfer coefficient prediction model for supercritical water based on Bayesian search algorithm

Abstract

In order to make better use of machine learning algorithm to perform thermo-hydraulic analysis of supercritical water reactor, different intelligent algorithms are used to optimise the model parameters for predicting the heat transfer coefficient of supercritical water. The accuracy changes of stochastic search and Bayesian search algorithms in predicting the heat transfer coefficient under different parameter spaces are compared and analysed. The results show that the search space and the initial distribution assumption have a large impact on the results. The Bayesian search algorithm is relatively less affected by the search space and parameter distribution assumptions. The prediction accuracy obtained by Bayesian search is 1.25–2.88% higher than that obtained by random search. After optimising the model parameters, the average test accuracy of predicting the heat transfer coefficient of supercritical water is more than 96.4%. At the same time, the spatial distribution characteristics of the optimal parameter points obtained by different search algorithms are analysed.