Transfer learning based fast generation method of practical dynamic security region boundary

• Transfer learning is introduced to rapidly generate practical dynamic security region (PDSR) for variable topologies. • Transfer learning leverages PDSR knowledge to derive security boundaries for topological faults, avoiding critical point recalculations. • Feature transfer with gradient reversal layers extracts cross-fault security patterns to generalize against renewable system fault diversity. • Test on the New England 39-bus system validates real-time security assessment, facilitating DSR integration in modern grids. The high proportion of renewable energy integration exacerbates the challenges of transient stability analysis of power systems. The practical dynamic security region (PDSR) based on the hyperplane form has outstanding advantages in situation awareness and a series of optimization problems. The key to generating PDSR lies in accurately and quickly obtaining a sufficient number of critical points located on the boundary. However, for different faults with changes in system topology, it is necessary to recalculate the critical points, which leads to the problem of not being able to quickly generate the security region boundary. Therefore, this paper proposes a rapid generation method for the PDSR boundary based on transfer learning. Firstly, feature transfer is used to minimize the distribution difference between the data of established faults and different fault operating points. On this basis, the gradient reversal layer is utilized to train and extract common features of established faults and different faults, update the parameters of the domain adversarial neural network model, and realize the identification of critical points for different faults. Finally, the PDSR boundary is generated based on the least squares fitting. The analysis of the New England 10-machine 39-bus system shows that for faults different from established ones but with similarities, the proposed method does not require recalculation. Based on the knowledge and experience of established faults, it accurately and quickly obtains the sufficient critical points needed to generate the boundary, significantly improving the generation speed of the security region boundary for different faults.