Research On Class Integration Test Order Generation Based On Evolutionary Optimization

Nowadays,information technology develops rapidly,and software has become indispensable in people's life.The quality of software has a significant impact on daily life and even the development of the society.People are increasingly dependent on software products and have higher requirements on software quality.For software manufacturers,low-quality software will lose its market competitiveness due to its high maintenance cost,while high-quality software can win the trust of consumers and thus bring higher economic benefits.Software testing became an important measure to ensure the quality of software.Through software testing,testers can timely detect software defects and submit for correction.In the process of software testing,integration testing is to test whether the modules can cooperate normally.Class is the unit of object-oriented program,and there are dependencies between classes.Different test orders have different test costs.Therefore,tester needs to determine a reasonable test order in order to reduce the test cost.Researchers have proposed different kinds of approaches for generating class integration test order.However,some approaches based on heuristic algorithm still have some shortcomings,such as the low optimization ability which leads to the relatively high test cost of generated class integration test order.In this thesis,some existing problems are optimized and improved,and two approaches for generating class integration test order are proposed.To solve the problem of blindness caused by random population initialization in Genetic Algorithm(GA)and Particle Swarm Optimization(PSO),this thesis introduces an approach for generating class integration test order based on initial population optimization.First,a constraint condition is introduced: strong dependencies between classes are not allowed to be broken.Then,a multi-way tree construction algorithm is proposed to form a multi-way tree forest.Finally,each multi-way tree in the forest is hierarchically traversed in random order to generate individuals that meet the constraint condition.The generated individuals form the initial population.The experiment results indicate that this method can generate the initial population with higher quality without losing randomness,leading GA and PSO to generate the class integration test order with lower test cost.PSO is easy to be precocious in the process of evolution.This thesis introducesan approach for generating class integration test order based on Dream Particle Swarm Optimization(DPSO).Each test order is considered to be a particle with the ability to dream.Each iteration cycle is divided into two phases: day and night.During the day,particles move from where they were at last night to new locations;At night,they contort the locations gained in the daytime according to their dreaming abilities.In this way,particles search near current locations,so that the algorithm can converge slowly and avoid falling into local optimum too early.The experiment results show that this method can generate class integration test order with lower test cost in a short time.In addition,EvoCITO,a class integration test order generation tool for Java programs is implemented,which adopts the above proposed approaches.