Boundary Scenario Generation Method of the Intelligence System Based on Adaptive Poisson Disk Sampling

Generation of boundary scenario is a common approach used to test and evaluate the intelligence system. It can generate amount number of test cases in regions that demonstrate critical transitions in performance modes. However, getting sufficient test cases within boundaries in a short time remains a huge challenge for a system under test. The choice of sampling method has a great influence on efficiency and accuracy of the test. This paper proposed a new black-box test method based on adaptive Poisson disk sampling to search the test cases in decision boundary of system under test. Firstly, the idea of sequential sampling and optimization are borrowed to guide Poisson disk sampling to sample in the boundary region of the black-box system. In order to improve the convergence speed of the algorithm, an adaptive search strategy based on dichotomous method is adopted. In addition, an elimination criterion is proposed to further decrease the calculation cost and the introduction of invalid samples. Finally, two state-of-art methods were compared on nine different categories of benchmarks and two actual intelligence systems to demonstrate the validation and efficiency of proposed method. The result shows that the proposed method can obtain more information of the system with fewer samples.