Public wireless communication networks enable power systems to manage demand-side resources (DSRs) efficiently. However, such networks are also vulnerable to random communication disruptions caused by cyberattacks or device faults. These disruptions may lead to the loss of control over large numbers of DSRs, thereby increasing the risk of frequency instability. To address this issue, this paper proposes a local-centralized collaborative control method for DSRs. First, the random characteristics of communication disruptions are incorporated into the frequency dynamic model of power systems with DSRs. Second, unlike existing research that relies on static local control, an adaptive stochastic local control strategy is introduced to guide the responses of DSRs when communication is unavailable. The aggregated behavior of these DSRs is then estimated and leveraged to centrally control the remaining DSRs, thereby avoiding uncoordinated or conflicting actions. Furthermore, a communication disruption threshold is defined to quantify the maximum level of disruption that does not impair the frequency performance. The effectiveness of the proposed method is validated through case studies under various disruption scenarios.
This letter reveals a fundamental limitation of passivity-based distributed stability analysis in power systems. Under the standard formulation, passivity certification inherently imposes a relative-degree compatibility constraint that excludes many high-fidelity inverter dynamic models (e.g., those that include electromagnetic transients). Potential extensions of passivity frameworks are discussed to break this limitation.
Hyperscale AI data centers have emerged as large, converter-dominated loads whose coordinated workloads challenge the long-standing assumption of load independence in power systems. Demand fluctuations at geographically dispersed substations can exhibit non-negligible temporal alignment that raises aggregate variability and undermines load diversity benefits in capacity planning. This paper presents an analytical approach to characterize spatial load correlation and identify the physical and operational mechanisms that drive synchronized behavior across distributed data-center facilities. The approach combines time-domain statistical measures and frequency-resolved spectral analysis to reveal how correlation emerges and propagates through the network. Real-time electromagnetic transient simulation on the IEEE 39-bus system validates the method against an independent Conventional Load baseline. AI data-center loads exhibit pronounced temporal non-stationarity. Correlated loading amplifies aggregate active-power variability about seven times above the independent reference and produces a frequency standard deviation eleven times larger. Low-frequency correlation reflects slow workload orchestration and thermal management dynamics. Higher-frequency alignment emerges from converter control interactions and network impedance characteristics. These findings establish spatial correlation as a characteristic structural feature of AI-dominated grids and motivate correlation-aware reserve sizing, interconnection screening, and operational frameworks.
The increasing integration of distributed energy resources (DER) offers new opportunities for distribution system operators (DSO) to improve network operation through flexibility services. To utilise flexible resources, various DER flexibility aggregation methods have been proposed, such as the concept of aggregated P-Q flexibility areas. Yet, many existing studies assume perfect coordination among DER and rely on single-phase power flow analysis, thus overlooking barriers to flexibility aggregation in real unbalanced systems. To quantify the impact of these barriers, this paper proposes a three-phase optimal power flow (OPF) framework for P-Q flexibility assessment, implemented as an open-source Julia tool 3FlexAnalyser.jl. The framework explicitly accounts for voltage unbalance and imperfect coordination among DER in low voltage (LV) distribution networks. Simulations on an illustrative 5- bus system and a real 221- bus LV network in the U.K. reveal that over 30% of the theoretical aggregated flexibility potential can be lost due to phase unbalance and lack of coordination across phases. These findings highlight the need for improved flexibility aggregation tools applicable to real unbalanced distribution networks.
The optimal deployment of switches and tie lines (TLs) plays a vital role in enhancing distribution network (DN) reliability. However, existing methods are primarily designed for radially operated DNs and rely on simplified reliability assessments, which renders them unsuitable for flexible inter-connected DNs (FIDNs) with a complex fault response process. To this end, this paper proposes an explicit reliability-integrated co-planning approach for multi-type switches and TLs in soft open point (SOP)-based FIDNs. First, a novel explicit reliability assessment method for FIDNs is proposed, considering the specific locations, operating features, and interdependencies of various types of switches and TLs, and fully capturing the coupled multi-stage fault response process. In particular, to enable the analytical formulation of FIDN topology, a node-branch state-association model integrating the SOP equivalent model is proposed. To accurately characterize the fault blocking and seamless transfer in FIDNs, a tight circuit breaker tripping model is established. Then, a mixed-integer linear programming (MILP)-based co-planning approach for multi-type switches and TLs in FIDNs is developed, which integrates the proposed explicit reliability assessment method. Case studies are carried out on a 37-node system, a real-world 304-node system, and a large-scale IEEE 8500-node system to validate the accuracy and integrability of the proposed reliability assessment method, and show that the proposed co-planning approach achieves lower annualized costs while enhancing system reliability.
The unit commitment (UC) problem is one of the key problems in short-term power system operation. By contrast, the clustered unit commitment (CUC) formulation is mainly used as a computational alternative to the UC model in long-term planning studies. Existing CUC methods introduce inherent errors that compromise feasibility and cost precision, particularly under scenarios with increased flexibility requirements. To overcome these limitations, this study presents two innovative models: Tight CUC (TCUC) and tight two-stage UC model (T-TSUC). Distinct from prior CUC methods, the TCUC model incorporates novel, strong, valid inequality constraints designed to tighten the formulation by explicitly addressing sources of inherent error, without introducing additional binary variables, which leads to a substantial decrease in computational time. The T-TSUC model employs a refined two-stage methodology: it first utilizes TCUC and then performs the UC in the second stage. This structure ensures feasible, detailed unit-level schedules and, in the studied cases, mitigates residual errors and reduces computational time while delivering near-optimal solutions. Numerical cases on three IEEE test systems under varying renewable penetration scenarios reveal that TCUC and T-TSUC outperform existing CUC methods in terms of computation time and accuracy in most cases, and in contrast to existing models, are less sensitive to the flexibility of the systems.
With the increasing demand for diverse test data in power system research, generating power grid models that satisfy power flow constraints have become a critical challenge. In this paper, we propose a progressive fine-tuning approach for large language models (LLMs) for the power grid model generation task. The initial phase applies supervised fine-tuning to enhance structural consistency and task adaptability, while the subsequent enhancement phase employs reinforcement learning with an expert knowledge–guided evaluation mechanism to ensure compliance with AC power flow constraints. To support fine-tuning, we construct a systematic dialogue-based dataset that encompasses label model generation and high-quality question–answer pairs construction, providing a framework that can be extended to other power system applications. Numerical experiments on generating power grid models of varying scales demonstrate that the proposed method significantly improves format accuracy, power flow residuals, and convergence rates, while also exhibiting strong generalization capability to previously unseen grid sizes. Its practical utility is further validated through a downstream Optimal Power Flow (OPF) learning task, where LLM-generated data serve as a benchmark to help identify the model with superior generalization capabilities.
With the widespread deployment of phasor measurement units (PMUs), data-driven composite load modeling has gained increasing attention among researchers. Existing approaches mainly rely on optimization-based methods to produce point estimates, which lack the capability to quantify estimation uncertainty. Alternatively, sampling-based techniques can provide confidence intervals (CIs) but are computationally expensive for real-time applications. In addition, the theoretical interpretability of such CIs is still limited. To address these issues, we propose a decentralized Bayesian load modeling strategy utilizing trajectory sensitivity that achieves structural decoupling between the load model and the external network. It eliminates the effects associated with external uncertainties in the system while effectively providing a probabilistic description of the load parameters, without resorting to time-consuming sampling. Moreover, for the first time we derive an analytical relation between trajectory sensitivities, measurement Jacobians, priors, noise, and confidence levels, theoretically demonstrating the rationale of key parameter screening in a statistical manner. The simulation results for multiple load modeling cases reveal the excellent performance of the proposed method.
In modern low inertia power systems with high inverter penetration, online inertia estimation has become challenging due to the presence of nonlinear converter control, virtual inertia loops, current saturation, measurement errors, and other practical uncertainties. To overcome these challenges, this paper presents an advanced Tree Pyramidal Adaptive Importance Sampling based Bayesian inference approach for the estimation of inertia and virtual inertia in an inverter-interfaced power system. Since the proposed algorithm possesses an adaptive proposal distribution refinement, it effectively explores the most probable parameter space. Hence, it reduces variance in inertia estimation results and converges early to the desired value of the inertia constant with fewer samples. The sample generation process is fully parallel, which supports faster execution. The proposed algorithm is robust against non-Gaussian posteriors that may arise due to non-linearity in the inertia estimation problem. The algorithm is hyperparameter-tuning-free and possesses an anytime property. The proposed algorithm is tested for inertia estimation in an IEEE 9 bus system interfaced with a PV-battery grid-following inverter for different scenarios of load shedding, load increment, weak and strong grids, non-Gaussian posterior, and Gaussian noise. Simulation and real-time results validate the superior performance of the proposed method for inertia estimation.
Extreme weather-driven cascading failures (CFs) in power systems can lead to catastrophic socioeconomic losses, yet integrating CF analysis directly into day-ahead proactive unit commitment (UC) is mathematically intractable. This difficulty arises from representing decision-dependent topology evolution within a tractable UC formulation. To address this challenge, a resilient UC framework is proposed, featuring a data-driven surrogate that encodes CF risks into a tractable polyhedral region. First, overload-induced CF simulations are performed offline across diverse commitment and dispatch states under contingencies. To capture nonlinear CF boundaries in the system's discrete-continuous operating space, a sequential training strategy is employed to construct a surrogate consisting of a series of linear classifiers. This surrogate identifies a polyhedral region with low CF risk, which can be directly integrated into the UC model as linear constraints for CF-aware decision-making. Case studies show that the proposed strategy significantly reduces the expected loss of load by proactively mitigating CF risks with only a marginal increase in operating costs, offering a streamlined yet effective pathway for enhancing power system resilience.
This paper introduces a novel approach for predicting system frequency response (SFR) and frequency nadir based on modal analysis. By decomposing the full system dynamic response, the method identifies dominant modes based on their participation in frequency behavior and derives a closed-form expression for the frequency trajectory. Unlike traditional approaches based on the Average System Frequency model, this method captures the true system dynamics and avoids oversimplified representations. The dominant modes exhibit low sensitivity to system parameters, enabling robust and accurate estimations across diverse operating conditions. The proposed approach is tested on three benchmark systems as well as the Salvadoran transmission planning network, demonstrating its scalability, precision, and adaptability. This methodology represents a shift from observing a simplified average system frequency response to a more detailed analysis focusing on system dynamics.
The power-flow constraints introduce strong power coupling of multiple virtual power plants (VPPs) in a distribution system, thereby increasing computational complexity and causing interdependent decision-making among VPPs. To address these challenges, this paper proposes a distribution operating security space (DOSS) based power decoupling mechanism of multiple VPPs. Independence, fairness, and customization are developed to characterize DOSS. First, a novel dual super-ellipsoid security region is constructed as a convex inner approximation of the power-flow space, ensuring distribution grid security while enabling customized DOSS formation. Second, the matrix decomposition, parametric programming, and constraint reasoning are employed to convert the DOSS's property descriptions into an optimization formulation. The convex quadratic DOSS derived from the optimization can be efficiently embedded into the aggregation and disaggregation of VPPs. It can satisfy the requirements of grid operational security and diversified independent operation of VPPs. Case studies show that the power transfers among multiple VPPs are effectively decoupled, and DOSS is equitably allocated in a differentiated scheme. Compared with other methods, the aggregated capacity of VPPs is significantly higher, and no power-flow violations occur during VPP operation.
Security-constrained unit commitment (SCUC) is among the important problems in the operation of power systems. Fast solution method for SCUC has long been a research topic that has attracted much attention. In this paper, a fast SCUC method is proposed based on an important finding. A mapping for each generator can be established, and this mapping is from a group of aggregated marginal costs to solution of individual generator. The mapping reflects the intrinsic property of a generator and is unrelated to the system to which the generator belongs. Based on this mapping, a historical-data-free and system-independent training of neural networks (NNs) could be implemented, which are employed to accelerate Lagrangian relaxation (LR) subproblems. Specifically, the system-level LR multipliers are updated by subgradients, and the binary solutions to individual unit subproblems are obtained directly by the trained NN. Thus, the computational burden for solving the subproblems and the dual problem is negligible. Numerical tests are implemented on IEEE 24-bus, IEEE 118-bus, and Polish 2383-bus systems, and the results show that the proposed method can give solutions of good quality in a short computational time.
Transmission planners require simple, transparent active distribution network equivalents (ADNEs) to accurately model distribution feeders with high penetration of converter-interfaced resources. Existing ADNE approaches rely on complex algorithms, hindering their routine adoption in transmission (bulk power system) planning studies; furthermore, validation has been confined to synthetic test networks. Thus, this letter presents an application-oriented procedure that reduces a detailed distribution feeder to a three-phase three-segment (TPTS) equivalent while preserving the original feeder characteristics required for transmission planning studies. The proposed approach has been validated on a real-world feeder in Arizona, modeled in DIgSILENT PowerFactory.
This letter proposes a highly accurate network-level rational time-domain model for gas-network transients. Traditional finite-difference and generalized-circuit methods often struggle to capture transient delay and wave-propagation characteristics without introducing numerical diffusion or response distortion. To address this issue, the exact Laplace-domain pipeline equations are first assembled into a global network transfer matrix, and rational approximation is then performed on the assembled network dynamics. This yields an explicit time-domain model that closely matches the fine-grid finite-difference benchmark and reproduces transient responses much more accurately than coarse discretization and generalized-circuit alternatives under realistic operating conditions.
The security region (SR) is an effective tool for assessing power system operational feasibility and security. However, frequent on/off adjustments of quick-start units in near real-time scale can break the fixed unit commitment assumption of traditional SR, leading to disjoint subregions and internal infeasible gaps. To address this, this letter proposes an iterative infeasible subregion identification approach, which integrates a two-stage optimization model with Farkas-lemma-based local exclusion cuts to iteratively detect and exclude infeasible subregions. Logical exclusion constraints are incorporated into subsequent iterations to prevent redundant searches until all disjoint regions are captured. Numerical studies demonstrate the method's effectiveness, precisely identifying internal infeasible regions, which occupy up to 12% of the convex relaxation in some cases, supporting secure and flexible operation.
This letter proposes an adaptive predispatch-hazard-redispatch (PHR) framework for multi-stage reserve dispatch in power systems. Unlike existing two-stage predispatch-redispatch and multi-stage heuristic threshold-based approaches, the PHR framework adopts adaptive redispatch weights to balance reserve adequacy assurance and computational cost. The complex multi-stage PHR structure is formulated as a policy graph and solved via a graph stochastic dual dynamic programming approach. Multiple strategies are explored to dynamically adjust redispatch weights based on system flexibility and uncertainty. Case studies on an IEEE 39-bus system demonstrate 70.0-86.5% cost reduction compared to benchmark methods while maintaining computational tractability.
This letter proposes a locally exact adjoint method to compute the locational marginal emissions (LMEs) in AC optimal power flow (ACOPF). Unlike conventional sensitivity analysis based on repeated load perturbations, the proposed method derives LMEs by linearizing the Karush-Kuhn-Tucker (KKT) conditions at the cost-minimizing equilibrium. We show that all-bus LMEs can be recovered simultaneously by solving a single reduced sparse linear system. Numerical results on the IEEE 118-bus and 2383-bus systems with heterogeneous thermal and renewable generation demonstrate that the proposed method captures clear spatial and temporal variation of LMEs effectively.
This letter presents an algorithm that efficiently and consistently represents, in the positive-sequence power flow, the phase shifts introduced by wye–delta-connected transformers. As is well known, wye–delta transformers impose fixed voltage and current phase shifts in multiples of 30 degrees. Because these shifts do not affect bus voltage magnitudes or branch power flows, they are frequently omitted or modeled inconsistently. However, non-uniform representation of transformer phase shifts can introduce artificial voltage-angle discontinuities that negatively impact Newton-based power flow convergence. To address this issue, the letter introduces the concept of Phase Shift Groups (PSGs), which enforce consistent voltage-angle representation across the power system model. The proposed PSG-aware initialization yields a physically consistent starting point that improves Newton-Raphson convergence robustness compared to conventional flat-start methods. Beyond initialization, PSGs provide a systematic mechanism for data-quality validation, flagging incomplete or inconsistent transformer connection data. Scalability is demonstrated on systems with up to 110,000 buses with negligible computational overhead.
