Research On Automatic Generation Of Test Data Based On Hybrid Dynamic Particle Swarm Optimization

Generating test data is one of hot points in the research field of software test. Particle Swarm Optimization, which has the characteristics of simple implementation, less parameters, fast convergence, automatic generation of test data quickly for the tested program, is vital to improve the efficiency of test and reduce test cost. For Particle Swarm Optimization exists the poor performance of local search and premature defects and low convergence precision, we study automatic generation of test data based on Particle swarm optimization, and improve a Hybrid Dynamic Particle Swarm Optimization to overcome the shortcomings in this thesis. the result of experiment shows the algorithm has the unique advantage through the program realization. The contents of this thesis are shown as follows:(1) For Particle Swarm Optimization exists the poor performance of local search and premature defects, An approach for generating structural test data based on modified the dispersion variation of simple particle swarm optimization using dynamically improving parameters is proposed. which according to branch path coverage of test guidelines. And taking into account the structural features of the branch predicate, a new fitness constructor function is introduced too.Through the open-test assemblies comparative tests confirmed, the result of experiment shows that the improved performance of the algorithm is better than the basic particle swarm optimization and parameters of the linear variation of particle swarm optimization from an average convergence path algebra and search time two aspects, when used to automated data generation.(2) Since the fully connected topology of particle swarm algorithm has low convergence precision and falls into local extremum easily, An approach for generating structural test data based on a hybrid particle swarm algorithm for automatically generating test data is proposed. Firstly, under the premise of global convergence, the population, which lack of diversity, uses Fixed-length ring topology to replace the fully connected one. Secondly, the roulette wheel method is introduced to select the candidate solutions and to update the location information and the velocity information. Lastly, for controlling and directing the particles to escape from local minimum, the conditions of tabu search algorithm is introduced too. The result of experiment shows that the hybrid algorithm has a better performance than the basic particle swarm optimization on population diversity. And the algorithm exhibited unique superiority by increased by 10% to 15% of the search success rate and path coverage on the basis of the genetic algorithm- particle swarm optimization algorithm in test data generation, while the average time-consuming is quite to the basic Particle Swarm Optimization.(3) Based on above research results, a Hybrid Dynamic Particle Swarm Optimization is improved, which is the organic combination of Hybrid Particle Swarm Optimization and Dynamically Parameters Particle Swarm Optimization. We use MATLAB language to demonstrate the implementation of the core program. Therefore, triangle determination program as a simple, demonstrate the efficiency of the path coverage, test case through comparing the three algorithms in generating test data, and analyzes the corresponding changes of the particle swarm diversity. Results show this method is practical and effective, which has superior performance.