A study on the development of digital model of digital twin in nuclear power plant based on a hybrid physics and data-driven approach

Abstract

The development of digital twin(DT) for nuclear power plants(NPPs) will further promote the use of nuclear energy. In the process of physical modeling for existing DT models, physical simulation modeling is not suitable for real-time monitoring and control of DTs. Moreover, most data-driven modeling methods with high computational efficiency do not consider the theoretical knowledge of the corresponding field, which will affect the application of research results in real situations. Therefore, to model the physical model of the DT model, this paper embeds physical knowledge into the backpropagation neural network(BPNN) to construct the data-driven model and physical-model combined neural network(DPNN) and further proposes the residual DPNN (ResDPNN) by introducing residual connections. This is a digital model developed by a hybrid physics and data-driven approach. Specifically, this paper embeds the law of energy conservation into the ResDPNN, constructs a high-precision digital model of the average temperature of the coolant using real operating data from a NPP, and conducts several sets of comparative experiments. The results show that the Mean Absolute Error(MAE) of DPNN on the testing set is improved by 28.02% compared to BPNN, and the MAE of ResDPNN on the testing set is improved by 7.66% compared to DPNN. The embedding of physical information and the introduction of residual connections effectively improve the generalization ability and predictive accuracy of the model. This can serve as a reference for the future development of DT for NPPs.