Fast calculation method of adaptive weights for water and steam physical properties based on neural network

The thermophysical property calculation module in the thermohydraulic safety analysis program of nuclear power plants greatly affects the calculation speed and accuracy of working conditions. Therefore, efficient algorithms for thermophysical properties are one of the key technologies for nuclear power software autonomy. To ensure the calculation accuracy of thermal hydraulic analysis software and improve the efficiency of physical property calculation, this study is based on the generalized regression neural network learning method and the multivariate interpolation method for efficient calculation of physical properties. We take the calculation of water and steam as an example. First, to improve the physical property form for the interpolation method, the neighborhood parameter memory algorithm is introduced for the interpolation calculation of largevolume data to simplify the calculation process. Then physical property parameters in the physical property database REFPROP get the accurate interpolation weights and establish the sampling scheme to generate the data set, and the interpolation calculation method is corrected by the predicted value of weight correction coefficients obtained by the training of the generalized regression neural network. After completing program development, we evaluate program performance by conducting a comparison of computational accuracy and efficiency. The research results show that improving the table and introducing the neighborhood memory algorithm can improve the efficiency of interpolation calculation, and the adaptive weight assignment module trained by coupled neural networks can effectively improve the accuracy of traditional interpolation calculation.