With the expansion of power systems, there is a significant increase in calculation quantity of the process of N-1 security check. In order to satisfy the increasing real-time requirements of N-1 security check, a graphics processing unit (GPU)-based acceleration algorithm of connectivity test in N-1 security check for large-scale systems was proposed. According to graph theory, the grid model was abstracted as a graph model, and then abstracted as a bio-directional vector graph model. Based on vector summation method, criterion of power grid connectivity was achieved in the vector graph model, and algorithm of connectivity test was designed. The case analysis shows that, the proposed algorithm has high efficiency and small memory footprint. It has significant advantages compared with traditional methods for connectivity test, which can effectively improve the speed of power grid connectivity test. Furthermore, it can help to improve the overall efficiency of N-1 security check in large-scale systems, and has engineering application value.