ABSTRACT This study investigates the improvement railway network Power Quality (PQ) parameters by optimal passive filter placement in railway traction networks. A MATLAB Simulink model of a traction network with fixed loading and three tuned filters (tuned to third, fifth, and seventh order harmonics) is developed. Simulations are conducted for each filter at different locations, and a mathematical model is created to evaluate filter efficiency. Results from both methods show that higher‐order filters (fifth and seventh) have negligible location‐dependent effects. Optimal filter placement is determined by simulating fifth and seventh order filters at the transformer location and varying the third harmonic filter location. The study evaluates PQ parameters, including current harmonics, voltage profile, and voltage harmonics, and finds significant improvement when filters are placed in the mid‐point of the traction network (section). In general, in the distribution network of power system filters are placed either at load end or source end based on the aim of the filter. In Railway networks this study becomes relevant because the load location is varying, impedance of traction line and transformer are in same range resulting in considerable line harmonic voltage drop. Moving the third harmonic filter to a 10–20 km range (away from traction transformer) improves the voltage drop by 2.3%, voltage THD by 2.4% at tail end which is within the recommended limits in IEEE 519–2022 and no major change in voltage drop, voltage harmonic and current harmonics at the source end.
ABSTRACT Security‐constrained unit commitment is a fundamental optimization problem in power system operations. Conditional transmission constraints, whose limit values dynamically depend on the number of online committed units, introduce computational challenges due to the binary variables required for level selection. This paper proposes a sparse formulation based on ordered binary variables to model these constraints. By representing the staircase function with an ordered binary sequence, the proposed formulation achieves the same linear programming relaxation tightness as the interval‐selection method while reducing binary variables and non‐zero entries. The ordered sequence associated with a given unit group is constructed only once and can be shared across multiple transmission sections controlled by the same committed‐number condition, which eliminates redundant auxiliary variables. Numerical results on three test systems show speedups over the conventional approach.
ABSTRACT Although islanded microgrids offer significant operational autonomy, their lack of external voltage and frequency references and absence of a global supervisory structure make them highly vulnerable to cyber‐attacks. Timely and accurate detection of such attacks is a critical prerequisite for maintaining microgrid stability and resilience. This paper presents a computationally efficient technique for real‐time detection and adaptive recovery of islanded AC microgrids against false data injection (FDI) and denial‐of‐service (DoS) attacks. Unlike complex machine learning‐ and observer‐based techniques, the proposed technique integrates a fourth‐order backward difference‐based predictor with the Teager–Kaiser energy operator (TKEO). Local frequency deviations are first estimated with high accuracy, generating a residual signal that is subsequently amplified by TKEO and assessed through an attack detection index (ADI). The key innovation lies in coupling this detection mechanism with an adaptive recovery scheme, which updates the communication adjacency matrix weights to isolate compromised links and restore network stability without prior knowledge of microgrid topology, attack type or location. Extensive simulations on an islanded IEEE 34‐bus test feeder demonstrate that the proposed technique achieves a high detection accuracy of 98.6% with a delay below 0.5 s. Moreover, it remains robust against measurement uncertainties, severe noise and typical microgrid disturbances such as sudden load changes and capacitor/reactor switching, while ensuring compatibility with diverse communication infrastructures. These results highlight the technique's effectiveness, resilience and practical applicability for next‐generation cyber‐secure microgrids.
ABSTRACT The rapid growth of large‐scale artificial intelligence (AI) data centres, with demand levels approaching to gigawatt (GW) scale, introduces significant challenges to transmission system voltage stability. This paper comprehensively investigates the impacts of AI data centres on static and large‐disturbance voltage stability in power systems. An adaptive continuation power flow, combined with contingency screening involving transmission lines and generator outages, is employed to quantify loadability limits and identify critical buses. Simulation results on the IEEE 39‐bus test system indicate that integrating a 1000 MW AI data centre load (equivalent to 10.84% of the system base load) substantially reduces the voltage stability margin from 36.15% to 5.35% under normal operating conditions. Under the most severe screened contingency, the margin is further reduced to 0.47%. Root mean square (RMS) time‐domain simulations further reveal that the case with the integrated AI data centre load exhibits poor post‐fault voltage recovery at the corresponding bus. The voltage converges to a low‐voltage equilibrium of 0.9584 pu, indicating insufficient reactive power support and weakened voltage restoration capability. To mitigate these adverse impacts, the voltage support functionality of battery energy storage systems (BESSs) is investigated. The default‐Q BESS case provides the highest updated static voltage stability margin, reaching 22.31% under normal operation and 18.50% under the most severe contingency. The full‐reactive‐capability BESS case provides margins of 18.71% and 14.16%, respectively, while giving the strongest RMS voltage recovery and increasing the post‐fault equilibrium voltage to 1.0582 pu.
ABSTRACT Faulted phase selection (FPS) is critical for the proper operation of protection systems. FPS is performed after fault detection, during which the relay observes signals whose characteristics vary depending on the fault detection time (FDT), including different levels of decaying induced transients mixed with the post‐fault cosine signal. However, numerous existing FPS methods overlook that fixed post‐fault sampling windows implicitly assume a known or fixed FDT and therefore do not address the variation or uncertainty in FDT encountered in relays. In this paper, it is demonstrated that the difference between the post‐fault and pre‐fault energy of voltage signals effectively represents the voltage drop component, based on which the voltage drop index (VDI) is proposed. Additionally, the induced transient index (ITI) is introduced based on the variance of the post‐fault voltage energy to estimate induced transients. By combining VDI and ITI, an adaptive fault classification index (AFCI) is proposed, in which the contribution of each index is adaptively adjusted using the Chatterjee correlation. By incorporating both aspects of fault characteristics, AFCI demonstrates the ability to classify faults across varying FDTs. The results indicate that the method accurately classifies faults across varying FDTs, even under noise and power swing conditions.
