Test Case Generation Method based on Generative Adversarial Network

Abstract

Traditional fuzzing tools generate low diversity of test cases and low vulnerability detection efficiency. This paper uses a test case generation model based on a generative confrontation network. The model uses LSTM as the generation network to generate data. The model uses fully connected network to construct a discriminant network for classification, and the test cases with the same protocol format are automatically generated after training. Finally the model uses the generated test cases to attack the protocol system to detect vulnerabilities. This method is compared with the traditional tools AFL, Peach and Sulley on the Modbus protocol, and the validity of the method is verified.