Research On Dynamic Multi-Objective Testing Resource Allocation

In the field of software engineering,software testing is one of the most effective ways to reduce system vulnerabilities,detect and correct errors.However,with the rapid development of computer technology,the scale of software systems is expanding and the system structure is becoming more complex.This results in that the invested testing resource and the consumed testing cost are also increasing significantly.Therefore,system reliability is no longer the only concern of software project managers.How to allocate limited testing resource and seek a satisfactory balance between system reliability,testing resource consumption and testing cost is a hot research topic in the field of software engineering in recent years.However,most of the existing research is limited to multi-objective static allocation of testing resource.In order to cope with the uncertainty of the system structure and testing resource caused by the change of the testing environment or user requirements in the testing phase,this dissertation mainly studies the dynamic multi-objective allocation problem of testing resource.The main research contents of this dissertation are as follows:(1)The research status and the practical applications of the testing resource allocation problem are analyzed.The multi-objective optimization model of the testing resource allocation problem is discussed,including the mathematical description of testing resources,system reliability and testing cost.This dissertation introduces the development history,principle,related technology and characteristics of multi-objective evolutionary algorithms,and focuses on two popular multi-objective evolutionary algorithms with superior performance: a fast and elitist multi-objective genetic algorithm NSGA-II and the third evolution step of generalized differential evolution GDE3.(2)In the optimization process,due to the change of user requirements,the number of modules or subsystems in the system structure may be increased.To tackle such situations,the corresponding population re-initialization strategy and constraint handling mechanism are designed,respectively.Then,a multi-objective testing resource allocation algorithm D-GDE3 based on GDE3 is proposed to deal with the dynamic change of the system structure.D-GDE3 considers system reliability,testing cost and consumed testing resource at the same time.It can reinitialize the population according to historical solutions to adapt to the change of system structure,and constrain new solutions to ensure feasibility.The comparative results show that D-GDE3 can converge quickly and obtain a solution set with better quality.(3)A mathematical model for multi-stage two-objective dynamic testing resource allocation problem is proposed,in which the total number of residual errors and the consumed amount of testing time are both minimized.Then,a multi-stage two-objective dynamic testing resource allocation algorithm MS-NSGA-II based on NSGA-II is proposed.Based on NSGA-II,MS-NSGA-II embeds parameter estimation,population re-initialization,and constraint handling to achieve multi-stage dynamic feedback and multi-objective optimization,and adaptively adjust testing resource allocation for each testing phase.The comparative results demonstrate that MS-NSGA-II can adapt well to the change of testing environments between each testing phase and provide more highquality testing resource allocation schemes.(4)A mathematical model of multi-stage tri-objective dynamic testing resource allocation is proposed to maximize system reliability,minimize testing cost and testing resource consumption.Then,a multi-stage tri-objective dynamic testing resource allocation algorithm MP-GDE3 based on GDE3 is presented.According to the constructed model,the corresponding parameter estimation,population re-initialization,and constraint handling are designed in the MP-GDE3 algorithm.Parameter estimation is an indispensable technology in multiple stages and the basis of software reliability growth model.Population re-initialization can save testing resource and make full use of historical information to obtain an initial population with good quality.Constraint handling can help individuals self-correct and accelerate the convergence speed of the solution set,thus obtaining a high-quality allocation scheme.The experimental results verify the effectiveness of the above improving strategies.