In recent years, the secure operation of power systems has grasped the focus of researchers' attention, and fuzzing test-based vulnerability mining has become a commonly used tool for evaluating its stability and robustness in the face of various anomalies. In fuzzing testing, test case generation is concerned with the ultimate performance of vulnerability mining, however, existing techniques generally focus only on coverage and not on the vulnerable parts. Therefore, based LSTM arrays that combined with residual connections, we propose a novel approach to learn the vulnerable parts of the power system to guide the process of fuzzing by utilizing the powerful feature representation capability of neural networks. Experiments show that the improved method can achieve faster convergence and higher vulnerability detection rate of neural networks, while the FNR of which is 20.4% and TPR is 79.6%.
|