ANN-based mathematical model for improving the accuracy of liquid flow measurements at nuclear power plants

Abstract

A review of literature sources demonstrates the relevance of improving the accuracy of liquid flow measurements. To solve this problem, a neural-network model for liquid flow determination was developed and tested. The optimum structure and training parameters of an artificial neural network, such as the activation function, transfer function of the output layer, number of hidden layers and neurons in them, were selected. The training sample was generated using empirical expressions of GOST 8.586.1–2005 (ISO 5167–1:2022). The developed neural-network predictive model, which provides an uncertainty of calculations no greater than 0.32%, is intended for use as part of a software and hardware system for improving the accuracy of liquid flow measurements at nuclear industry enterprises.