DEVELOPMENT AND EVALUATION OF FLOW REGIME IDENTIFICATION MODEL BASED ON MACHINE LEARNING ALGORITHM

Abstract

The accurate identification of the flow regime is important for two-phase simulation in nuclear reactor design and analysis. In this paper, the flow regime identification model was developed based on the decision tree (DT) algorithm, random forest (RF) algorithm, and artificial neural network (ANN) respectively. Training data was collected from available literature and classified into nine flow regimes. The input parameters of the flow regime identification model were superficial liquid velocity, superficial gas velocity, hydraulic diameter, pressure, and flow direction. Then the three flow regime identification models were evaluated by precision, recall, and F-score which is the composite value of precision and recall. Moreover, the computation time of the three models was also compared. The results show that flow regime identification models based on the RF algorithm and DT algorithm have better accuracy than ANN. The accuracy of the model based on RF and Dt is 94.2% and 97.0% respectively while the accuracy of the model based on ANN is 77.0%. What’s more. the models based on the DT algorithm and ANN have less time-costing than the RF algorithm. The predicting time cost of the models based on the DT algorithm and ANN is 0.47s and 0.74s respectively while the predicting time cost of the model based on RF is 47.0s.