Study On Fuel Loading Pattern Optimization For A Pressurized Water Reactor Based On NSGA-? And Machine Learning

The quality of the PWR fuel loading scheme is related to the economic and safety of the nuclear power plant directly.In this paper,two main problems in the optimization of PWR fuel assembly loading schemes are discussed:large feasible solution set in the problem and long assessment of fuel loading scheme.Due to time consuming of solving neutron transport equation,this paper explores a faster way to solve neutron physics calculation.Some kinds of machine learning algorithms are proposed to fit the burnup calculation,infinite multiplication factor and power peak factor for 3×3,5×5,7×7,9×9,14×14,17×17 fuel assemblies.Also the feature importance of fuel loading problem are analysed based on the tree model.The study found that under the condition of large sample size,the fitting ability of all kinds of machine learning models is good,and the fitting error is acceptable.Lightgbm model is relatively optimal among all of models.The loading of ²³⁵U in the assembly has a great influence on the fitting effect of various neutron physical parameters.The bias value of assembly loading scheme has a great influence on the fitting effect of power peak factor and infinite proliferation factor.After fine-tune modes,the neural network algorithm and the lightgbm model have a good ability to predict the neutron physical parameters of the 17×17 fuel assembly,which verifies the feasibility of the machine learning algorithm to evaluate the fuel assembly loading scheme.By using the NSGA-? algorithm and the machine learning model,this chapter attempts to solve the problem of large search space and long evaluation time in fuel loading optimization problem.In this chapter,SiC cladding accident-tolerant fuel is selected as the research object.After exploring the neutron physical feasibility and advantages and disadvantages of SiC cladding accident-tolerant fuel,the SiC cladding accident-tolerant fuel assembly is used by NSGA-? algorithm and machine learning model for optimization research.Under the condition of a relatively small number of samples,the model has a better fitting effect on the infinite multiplication factor then power peak factor.In the meantime,error is in an acceptable range.Applying machine learning for evaluation of the loading scheme in the assembly optimization process,the optimization loading scheme can be obtained in a quite shorter time than regular deterministic method.Confirming the feasibility of using machine learning to evaluate the loading scheme during assembly loading scheme optimization.