Abstract
The methodology of artificial intelligence (AI), particularly artificial neural network (ANN), would be in favor of nuclear energy system development. These ANN simulators may provide more efficient means than the traditional nuclear design codes, especially for the design of the key parameters, such as core geometry and layout, material composition.
In this paper, a neutronics calculation code SARAX and the corresponding multilayer perceptron (MLP) surrogate model were used as simulators for core parameters optimization of a reference lead based fast reactor. The pellet radius, enrichments and active height are the interested core parameters, and core burnup and power distribution are the target characteristics in the study. The training of structure and weight parameters in MLP network are based on about 5000 calculations of SARAX code. Test results of neural network show a good agreement between MLP surrogate model and SARAX code. The feasibility of the MLP surrogate model to be used in core parameters optimization was also discussed. Results showed that, the core surrogate model based on MLP could be quickly constructed and regulated, and be a more efficient simulators in a innovate reactor optimization.
The above work is completed in Sinan Platform, a multidisciplinary intelligent design platform developed by China Nuclear Power Technology Research Institute Co. Ltd.
