Dynamic security assessment of power systems using a deep learning and GAIN-based approach for addressing missing data

• Robust Handling of Missing Data: The framework effectively manages simultaneous missing data from multiple PMUs over consecutive time intervals, maintaining high DSA accuracy even under substantial data loss. • Topology-Independent Data Recovery: The GAIN algorithm’s ability to impute missing data without relying on system topology enhances its applicability across diverse power system configurations, offering a significant improvement over topology-dependent methods. • High Predictive Accuracy with CNN : The CNN-based DSA model extracts critical features from high-dimensional PMU data, outperforming simpler ML models like SVM by generating robust and informative features through convolution and pooling operations. • Scalability and Versatility: Extensive validation on the IEEE 39-bus and IEEE 118-bus systems demonstrates the framework’s scalability, ensuring reliable performance across a wide range of system sizes and operating conditions. • Comprehensive Sensitivity Analysis: A rigorous sensitivity analysis of the GAIN algorithm under varying levels of missing data confirms its robustness, providing confidence in its real-time applicability for DSA. The growth of artificial intelligence tools has sparked a surge in research attention to data-driven power system dynamic security assessment (DSA). Despite these advancements, DSA approaches face substantial challenges, including unintended data losses across communication, processing, and storage stages, as well as intentional data disruptions caused by cyberattacks. Such data issues can impair the effectiveness of DSA techniques and lead to erroneous decisions. This research introduces a novel hybrid data-driven approach for restoring missing data in phasor measurement units (PMUs) by the generative adversarial imputation network (GAIN) algorithm. It utilizes advancements in deep learning (DL) to develop a convolutional neural network (CNN) model for accurately assessing the dynamic security of power systems. The proposed model has been validated using the IEEE 39-bus and 118-bus networks, and the outcomes demonstrate the high accuracy and robustness of the proposed algorithm in achieving up to 99.48 % classification accuracy and maintaining stable performance even with up to 50 % missing data.