A Multi-Objective Generative Graph Transformer for Co-Optimized Transient Stability in IBR Grids

Inverter-Based Resources (IBRs) have rapidly lowered current power grids rotational inertia, necessitating quick dynamic Transient Stability Assessment (TSA) and mitigation. Deep learning has improved data-driven TSA, but several gaps remain: accuracy reductions with missing PMU data, have trouble generalizing over dynamic topological changes, and segregate stability evaluation from active control mechanisms. The Multi-Objective Active Spatio-Temporal Graph Transformer (MO-ASTGT) is a deep learning system for transient stability prediction, data reconstruction, and grid management.A Conditional Tabular Generative Adversarial Network (CTGAN) correctly imputes missing or damaged PMU data streams, ensuring ongoing observability. Sliding Window Dynamic Graph Attention (SW-DGA) and a Multi-Head Transformer map high-IBR grid dynamic topological changes to capture complex, long-range spatio-temporal connections. A Domain Adversarial Neural Network (DANN) with active transfer learning enable good performance under unanticipated errors. MO-ASTGT abandons sequential assessment-then-control. It provides a multi-objective control vector that reduces Emergency Load Shedding (ELS) and maximizes VSG virtual inertia and FACTS support.The framework is modeled and verified in using PyTorch and an IBR-augmented IEEE benchmark test system, independent of older simulation tools. Performance comparisons with ST-GHAN, DANN-FMC-IAL Bi-LSTM, and hybrid GCN-Transformers indicate operational superiority. Fixed topologies and comprehensive data observability give MO-ASTGT 99.85% prediction accuracy. The suggested model achieves over 96.8% accuracy under extreme informational strain, including a 30% random loss of PMU data and topological changes, yielding 16.2% better predictive resilience than hybrid models. The unified multi-objective control technique reduces absolute megawatt load shedding by 25.4% compared to deterministic, threshold-based ELS systems by dampening rotor angle oscillations.MOSTGT transforms transient stability from a passive diagnostic tool to a proactive, robust control mechanism. This work combines sophisticated graph-based generative deep learning with continuous grid augmentation in a deployable, open-source-ready architecture to provide a viable, scalable Energy Management System (EMS) solution for next-generation smart grids.