With the construction of large-capacity long-distance high-voltage direct current transmission projects and the large-scale integration of renewable energy, frequency security of the power system is facing severe challenges. For fast and accurate online assessment of frequency security, a data-driven model based on a metric learning (ML) and generative adversarial network (GAN) is proposed. First, the key frequency security indicators are selected as the outputs of the model, and an input feature set is constructed. Then, the improved Wasserstein generative adversarial network (WGAN) based on the Wasserstein distance metric is used to learn the distribution information of historical operation scenarios of power systems. This generates operational scenarios covering typical modes to build the training sample set. Considering the inapplicability of a single machine learning model to frequency security assessment with complicated operational modes of power systems, a combined assessment model for assessment composed of multiple sub-models is constructed based on metric learning for a kernel regression (MLKR) method. Finally, a simplified Shandong power system example is used to verify the effectiveness of the proposed method.