Research On Mining Metamorphic Relations Of Burnup Calculation Program Based On Regression Analysis

Verification and validation(V & V)of nuclear power software is an important means to ensure the correctness and reliability of numerical simulation software for complex systems and ensure software quality.The verification of autonomous nuclear power software mainly adopts the direct comparison method.There are four common methods for expected results: benchmark question,power plant operation data,experimental data,and classical program output.Because the mathematical and physical equations solved by nuclear power software are mostly partial differential equations,it is usually difficult to construct analytical or accurate solutions,and the development cost of benchmark problems and experimental data is high and the cycle is long.There is almost no reference program for advanced reactor software and new generation nuclear power related software.The software verification faces the problem of difficult determination of calculation results,which makes the verification insufficient and affects the safety and economy of engineering analysis and design.That is,the verification and validation of nuclear power software has test oracle problem.Metamorphic testing is an indirect comparison technology and one of the mainstream and important technologies to alleviate the above problems of Oracle.It indirectly evaluates the correctness of the program by checking whether the input/output meets the metamorphic relation without constructing verification examples.It has been preliminarily and successfully applied to the verification of a number of domestic autonomous nuclear power design software and achieved good results.Metamorphic relation is the core of metamorphic testing.At present,the research on metamorphic testing mainly focuses on the identification method of metamorphic relation.Therefore,this paper takes the burnup calculation program as the research object,and proposes a framework of likely metamorphic relation identification,which includes three stages: input pattern identification,output pattern mining and metamorphic relation verification.The research contents of this paper mainly include:(1)For the research of input pattern in the burnup program,the input pattern identification rules based on domain knowledge is established.Based on the background knowledge of the program to be tested,the rules are established by analyzing its mathematical and physical equations,numerical solution algorithm and program code,and through the data variation and combination.Based on the guidance of domain know-ledge,more meaningful input patterns can be identified through the static analysis,which is of great significance to the mining of next output patterns.(2)For the research of output pattern in burnup program,the method of metamorphic relation output pattern identification based on regression analysis is proposed.The test input pairs are generated according to the input patterns by using the random value generation strategy,and the corresponding output data pairs are obtained by executing the program.Then using an appropriate regression model to mine the output patterns in the burnup calculation program.(3)For the burnup calculation program--NUIT,the empirical study of likely metamorphic relation mining between the power and nuclide density of NUIT,a number of valuable metamorphic relations are identified,and the effectiveness of the method is verified.In this paper,the metamorphic testing technology is introduced into the verification of nuclear power software.The above research shows that the proposed metamorphic relation mining method based on regression analysis is effective.This method reduce the difficulty of finding the metamorphic relation,and save human and material resources.