Ammonia (NH3) co-firing is a practical option for reducing direct CO2 emissions from coal-fired power plants while utilizing existing infrastructure. However, its overall environmental and economic feasibility depends strongly on the NH3 supply pathway and policy support. This study evaluated the process performance, cradle-to-gate environmental impacts, and economic feasibility of NH3 co-firing in a 1,100 MW ultra-supercritical circulating fluidized bed (USC-CFB) coal-fired power plant. The NH3 co-firing ratio was varied from 0 % to 50 % (HHV basis), and NH3 was supplied from fossil-based (grey), fossil-based with carbon capture (blue), and renewable (green) production routes. Life cycle assessment, techno-economic analysis, policy scenario analysis, sensitivity analysis, and Monte Carlo-based uncertainty analysis were conducted. Increasing the NH3 co-firing ratio reduced direct CO2 emissions at the power generation stage but decreased gross power output and net plant efficiency. At 50 % NH3 co-firing, green NH3 reduced global warming potential by 33.1 % relative to baseline case, whereas grey and blue NH3 increased by 18.1 % and 2.0 %, respectively. The levelized cost of electricity (LCOE) increased with NH3 co-firing ratio for all pathways in the order of grey < blue < green, mainly because of NH3 fuel cost. With policy support, the LCOE of green NH3 could be reduced to below 100 $/MWh in the range of 20–40 % NH3 co-firing ratios. Overall, the combination of green NH3, intermediate co-firing ratios, and policy support provided the most favorable balance between GWP reduction and economic performance among the evaluated cases.
Large Language Models (LLMs) are increasingly applied to cybersecurity, but their reliability in Critical Infrastructure (CI) threat modeling remains underexplored. This study proposes CITADEL, a CTI-grounded framework for generating CI-specific cyber threat scenarios with open-source LLMs. CITADEL combines MITRE ATT&CK, CAPEC, and CISA KEV in a three-stage Planner-Enricher-Writer pipeline to produce structured, evidence-linked multi-stage scenarios. We evaluated four open-source LLMs across Communication, Energy, and Healthcare domains through a blind assessment by ten cybersecurity professionals. The results indicate generally positive expert ratings for realism, accuracy, coherence, and utility, while revealing domain-specific differences in model performance. Compared with unconstrained LLM generation, CITADEL improved ATT&CK technique identifier validity by 44.9 percentage points and XML parse success by 7.1 percentage points. CAPEC semantic accuracy also increased by 33.8%, though this improvement did not reach statistical significance and should be interpreted as a directional finding. These results suggest that CTI grounding substantially improves the factual grounding of attack technique identifiers and the structural reliability of generated threat scenarios. Overall, CITADEL provides a practical blueprint for evidence-grounded LLM-based threat scenario generation in CI security.
In this paper, we investigate the cardinality of the distance set Δ(A,B) for sets A and B⊂Fqd where A or B is contained in a k-dimensional affine subspace over a finite field. Assuming that B lies in a k-coordinate plane up to translations and rotations, we prove that if |A||B|>2qd, then |Δ(A,B)|>q2, where |Δ(A,B)| denotes the number of distinct distances between elements of A and B. In particular, we show that our result recovers the sharp (d+1)/2 threshold for the Erdős–Falconer distance problem in odd dimensions, where distances are determined by a single set. As a further application, we also obtain an improved result on the Box distance problem posed by Borges, Iosevich, and Ou, in the case where 2 is a square in Fq.
In the era of Industrial Internet of Things (IIoT) and Cyber-Physical Systems (CPS), Federated Learning (FL) offers a promising decentralized intelligence paradigm for Video Anomaly Recognition (VAR). This task is vital for maintaining high-fidelity Digital Twins and ensuring safety in mission-critical environments. However, the inherent data heterogeneity across distributed edge clients leads to a fundamental challenge known as semantic misalignment, where clients learn divergent feature representations of “normal” and “abnormal” events. The problem becomes particularly pronounced in VAR, where the presence of diverse and fine-grained anomaly categories leads each client to develop distinct semantic interpretations of abnormality. Existing federated methods primarily focus on binary anomaly detection and fail to address this misalignment, preventing effective fine-grained recognition. In this paper, we introduce FedVAR, a weakly-supervised FL framework explicitly designed for VAR. Leveraging the rich representations of Vision-Language Models (VLMs), FedVAR employs a prototype-based alignment mechanism that creates a shared semantic anchor for all clients to re-center and align their visual and textual feature spaces. This process enforces a consistent representation of “normality” across the decentralized network, directly mitigating semantic misalignment and enabling robust prompt-learning of anomaly direction vectors with minimal communication overhead. We conduct extensive experiments on challenging benchmarks under various non-IID data partitioning schemes, unseen domains, and novel anomaly classes. The results demonstrate that FedVAR consistently outperforms state-of-the-art federated baselines, establishing a robust framework for distributed intelligence in video-based CPS.
