The rapid deployment of electric vehicle charging stations (EVCS) within distribution networks requires intelligent, adaptive control to maintain the grid's resilience and reliability. In this work, we propose PHANTOM, a physics-aware adversarial network through training and optimization of multi-agent reinforcement learning model. PHANTOM integrates a physics-informed neural network (PINN) enabled by federated learning (FL) that functions as a digital twin of EVCS-integrated systems, ensuring physically consistent modeling of operational dynamics and constraints. Building on this digital twin, we construct a multi-agent RL environment that uses deep Q-networks (DQN) and soft actor-critic (SAC) methods to develop adversarial false data injection (FDI) strategies that can bypass conventional detection mechanisms. To examine the broader grid-level consequences, a transmission-distribution (T&D) dual simulation platform is developed, allowing us to capture cascading interactions between EVCS disturbances at the distribution level and the operations of the bulk transmission system. Results demonstrate how learned attack policies disrupt load balancing and induce voltage instabilities that propagate across T&D boundaries. These findings highlight the critical need for physics-aware cybersecurity to ensure the resilience of large-scale vehicle-grid integration.

Incorporating advanced information and communication technologies enhances smart grid (SG) operation, while increasing vulnerability to false data injection (FDI) attacks. Identifying and characterizing FDI attack vectors is crucial, as they can jeopardize SG system stability and protection. State-of-the-art (SOTA) attack analytics predominantly employ machine learning (ML) to extract attack vectors that can evade rules-based bad-data detectors. While scalable, these approaches offer no guarantees of identification or stealth and often yield simplistic attack vectors detectable by ML-based anomaly detection models (ADMs). Formal methods, in contrast, can synthesize verifiable attack vectors while ignoring ML-based ADM. Several tools in other domains attempt to identify attack vectors against ML-based ADMs; however, they apply to systems with straightforward control dynamics and cannot be directly transferred to complex, interdependent SG control systems. To address these gaps, we introduce MISGUIDE, a defense-aware attack analytics that jointly models LFC dynamics and an ML-based ADM to extract verifiable, multi-timeslot FDI attack vectors that can trip under/over-frequency relays while remaining stealthy. The ADM used in MISGUIDE can detect 100% of the attack vectors found by SOTA attack analytics. Using real-world load data, we validate the attack vectors generated by MISGUIDE through hardware-in-the-loop OPAL-RT simulations on the IEEE 39-bus system.

As the distribution systems (DSs) are transitioning into active distribution networks (ADNs) with increasing photovoltaic (PV) penetration, computationally tractable dynamic models become necessary for real-time analysis and control. This article presents a two-stage data-driven modeling framework based on NARX type recurrent neural networks (RNNs) to develop dynamic equivalent models (DEMs) for smart PV systems and PV-integrated ADNs. Initially, DEMs are developed to capture the nonlinear power dynamics of residential PV systems with ancillary voltage support. The framework is further extended to autonomously capture substation-level ADN power exchange across varying PV penetration levels. The practical implementation of the DEMs is supported by integration strategy that allows seamless coupling of the data-driven models with conventional modeling counterparts. The proposed approach is validated on modified IEEE 123-node and 8500-node test feeders under unbalanced loading and dynamic environmental profiles. Case studies on 123-node feeder show that the RNN-based DEM of ADN achieves average MSEs of$4.12\times 10^{-4}$p.u. for active power and$7.04\times 10^{-4}$p.u. for reactive power, while maintaining robustness across PV penetration levels up to 60%. Importantly, the proposed DEM provides over$1000\times$simulation time speed-up compared to detailed model-based PV-integrated ADNs for the larger test feeder, offering a scalable solution for both DSOs and TSOs in dynamic power flow studies.