An Improved Many-objective Optimization Algorithm MOEA/D-VW And Its Application In Test Case Prioritization

Multi-objective optimization problems(MOPs)are a very common problem in the real world.Multiple objectives of MOPs need to be optimized simultaneously,while these objectives often conflict and cannot reach the optimum at the same time.Multiobjective optimization algorithms are effective methods in solve MOP,in which decomposition based multi-objective optimization evolutionary(MOEA/D)algorithms have become one of the most widely developed and most significant algorithms because of its strong search ability,high robustness,and high expansibility.However,existing MOEA/D algorithms still have the following problems: 1)It is unable to deal with complex Pareto frontier problems;2)When dealing with many-objective optimization problem(Ma OPs),the distribution of obtained solution sets is uneven,and the overall performance of MOEA/D is poor;3)The time cost of many MOEA/D variants is significant.In this thesis,we attempt to improve the MOEA/D algorithm and apply the improved algorithm to resolve the test case prioritization(TCP)problem.The specific research content is as follows:(1)In order to overcome the existing problems in MOEA/D,this thesis proposes a multi-objective optimization algorithm MOEA/D-VW based on adaptive weight vectors.This algorithm adopts a lightweight adaptive weight vector updating strategy to adaptively adjust the weight vectors to guide the evolution direction of the population.At the same time,by improving the crossover operator,the population can have a wider search space in the early stage and can quickly converge in the late stage of the search.Also,it is not easy to fall into the local optimum.In addition,updating the neighborhood periodically can improve the performance of the algorithm and reduce the time cost.In this thesis,MOEA/D-VW is compared with 7 advanced multi-objective optimization algorithms on 31 groups of test functions.Experimental results demonstrate that MOEA/D-VW can effectively improve the distribution of solution set,and convergence and robustness of the algorithm.(2)Test case prioritization is an important task in software engineering.It aims to help developers to reveal the defects in software as quickly as possible by prioritizing test cases.This thesis discretizes MOEA/D-VW into algorithm PMOEA/D-VW and uses the decision points of MOEA/D as the low-level heuristic algorithm to construct a super-heuristic algorithm HHMOEA/D,then applies PMOEA/D-VW and HHMOEA/D to the test case prioritization problem.In this thesis,we construct code coverage matrices for six projects and analyze the performance of PMOEA/D-VW with different combinations of objectives.Experimental results show that PMOEA/D-VWbased on APDC+EET has the best performance in test case prioritization.In addition,PMOEA/D-VW is compared with 8 test case prioritization strategies.Experimental results show the superiority of PMOEA/D-VW in test case prioritization.HHMOEA/D is independently executed 30 times on TCP problems of 6 projects.Experimental results demonstrate that HHMOEA/D can effectively select the most appropriate low-level MOEA/D algorithm to solve the TCP problem of the target project.