Prediction of steam/water stratified flow characteristics in NPPs transients using SVM learning algorithm with combination of thermal-hydraulic model and new data mapping technique

Abstract

Steam/water stratified flow would occur in transient condition (e.g. LOCA) in light water Nuclear Power Plants (NPPs). Due to high gradient of flow characteristics at the interface of steam/water flow, the prediction of flow characteristics (e.g. temperature, pressure, velocity, and Turbulent Kinetic Energy (TKE)) requires further attention and special interfacial models. Also, accurate simulation of these mentioned characteristics needs fine spatial mesh and very small time steps based on Computational Fluid Dynamics (CFD) standard criteria. In order to reduce the computational cost, the combination of thermal–hydraulic modelling and soft computing is considered as a new strategy in this study. The steam/water stratified flow in a rectangular channel (case 3 of Lim et al test section) is examined as case study and calculated values of the characteristics by thermal–hydraulic model are fed as training/test data to the Support Vector Machine (SVM) learning algorithm. SVM in combination with the proposed data mapping technique which is a type of autocorrelation finding predicts the value of each characteristic at a specific position/ time using the value of that characteristic at previous time at that position and previous position. The results show that the proposed methodology is appropriate for prediction of steam/water flow characteristics. Velocity, temperature, and TKE are predicted with reasonable accuracy. The predicted pressure shows a trend similar to the values obtained from the thermal–hydraulic modelling. For precise prediction of parameters similar to the pressure, it seems deep learning in combination with the proposed data mapping technique and a kind of features selection technique are needed. This method is under development and will be reported as the subsequent.