Prediction Of Neutron Flux Density And Effective Increment Coefficient In PWR Core Based On Deep Learning

With the continuous development of reactor engineering technology,the requirements for reactor physics calculation are increasing.The traditional neutron diffusion equation has some problems,such as long calculation time and large storage space.Therefore,it is of great significance to study how to obtain the diffusion equation results quickly and accurately.With the continuous improvement and development of modern computer technology and performance,as well as the rapid development of modern artificial intelligence,the performance of computer is enough to maintain the high-performance computing urgently needed by modern artificial intelligence research,and artificial intelligence has also been developed rapidly,in which machine learning has become an important means to solve some practical problems.In the physical analysis of nuclear reactor core,for the solution of diffusion equation,due to its complex structure and multi system coupling characteristics,the traditional numerical calculation method has the disadvantages of long calculation time and large storage space requirements.In view of these shortcomings of traditional methods,this paper proposes a deep learning model to predict the core parameters.Firstly,the traditional numerical analysis method is used to establish the core numerical simulation,and the more commonly used difference method is used to calculate a large number of nuclear reactor core numerical simulation data through the original iteration method and over relaxation iteration,Then BP neural network and convolution neural network are used to build the deep learning model which can map the relationship between the core structure parameters and the core parameters.For keff,BP neural network and convolution neural network are used to predict the neutron flux density,and single channel and multi-channel convolution neural network are used to establish the prediction model.In this paper,Python code is used,Tensorflow framework for in-depth learning.Finally,a deep learning model is obtained to predict the core parameters of specific reactor type.The results show that:in terms of keff prediction,MSE of convolution neural network and BP neural network prediction models are 2.54×10^-4and 8.14×10^-4respectively,and the average relative errors of BP and convolution neural network prediction models for test samples are 0.042332 and 0.020799 respectively.Therefore,the prediction accuracy of convolution neural network is better than that of BP neural network.In the prediction of neutron flux density,the MSE of multi-channel convolution neural network and single channel convolution neural network are 6.088×10^-6and7.328×10^-6respectively.Therefore,the prediction accuracy of multi-channel convolution neural network and single channel convolution neural network are in the same order of magnitude,and there is no significant difference,but the multi-channel convolution neural network is more universal in the prediction of core parameters.Both BP and convolution neural network can predict the core parameters with a single sample time of about 0.25 Ms.compared with the traditional numerical simulation method,its prediction efficiency is very high.