Abstract

The rapid growth of data center energy demand has intensified the need for efficient and low-carbon energy supply. While SOFC-based combined cooling, heating, and power (CCHP) systems and data center cooling flexibility have been widely studied, their joint optimization within a unified framework remains limited. This study develops an SOFC-based CCHP model that integrates electricity supply, waste-heat-driven cooling, and cooling flexibility into a coordinated optimization framework. The results show that SOFC-integrated coordinated electricity–cooling operation can reduce operating cost and grid dependency while improving energy utilization.

Abstract

The high energy penalty of chemical-absorption carbon capture limits its large-scale deployment. Overhead gas recompression (OGR) reduces this penalty; however, conventional OGR generates high interstage temperatures, and applying deep inter-cooling causes latent heat loss, diminishing recovery benefits. To address this, a continuous in-stage spray cooling process using downstream condensate is designed to lower temperatures and maximize heat recovery. Because the spray reflux alters the fluid mass across compression stages, conventional equal-pressure-ratio strategies are inapplicable, requiring multi-dimensional optimization of pressure ratios and spray diversion ratios. Since mechanistic optimization of flowsheets with complex recycle loops is computationally prohibitive, a boundary-sensitivity analysis is first conducted to thermodynamically decouple the OGR module from the upstream absorption-desorption process. Using Latin Hypercube Sampling and automated Aspen Plus simulations, a dataset is generated to train a deep neural network (DNN) surrogate model. Coupled with the NSGA-II algorithm under a thermo-mechanical safety limit, the framework optimizes specific heat consumption, specific power consumption, and total annualized cost (TAC) to determine the optimal parameters. Finally, SHAP and thermodynamic path analyses are applied to interpret the black-box model, confirming that the optimized wet compression balances evaporative cooling and mass-flow penalties. Compared to the conventional OGR baseline, the optimized configuration reduces specific power consumption by 12.7% (to 131.9 kWh/tCO2) and lowers the TAC by 3.6% (to 87.80 million USD/yr), increasing the equivalent coefficient of performance (COPeq) from 6.13 to 9.94.

Abstract

Centrifugal compressors are widely used in the energy industry due to their simple structure, stable operation, high pressure ratio, and excellent efficiency. This study presents a multidisciplinary optimization framework integrating computational fluid dynamics with hybrid modeling for centrifugal compressor impeller design. To address limited sample challenges, we develop an improved artificial neural network enhanced by neighborhood component analysis for dimensionality reduction and genetic algorithm optimization, achieving 42% faster convergence than conventional methods. Our multi-objective approach simultaneously maximizes isentropic efficiency and operational stability across off-design conditions. Experimental validation demonstrates: (1) 1% improvement in design-point isentropic efficiency, and (2) 0.3% wider high-efficiency operational range compared to baseline designs. Centrifugal compressors account for more than 30% of the power consumption in industrial production and energy systems. Improving their efficiency and stability plays a significant role in reducing energy consumption and enhancing overall efficiency.

Abstract

Multi-stack solid oxide fuel cell (SOFC)systems suffer from issues such as state of health (SOH) heterogeneity, migration of optimal power regions caused by degradation, and high computational complexity in high?dimensional optimization during long-term operation. Based on the power evolution characteristics of a single stack under varying SOH, this paper proposes the concept of power equivalence and establishes its
quantitative model. Within the framework of approximate dynamic programming (ADP), the power
equivalence factor is employed as a state aggregation operator to map the high-dimensional SOH space to low?dimensional reference states, thereby reducing the complexity of lifecycle optimization. Meanwhile, a SOH?dependent safe power upper bound constraint is introduced to determine the feasible operating region of each stack. The above work lays a critical theoretical and methodological foundation for the further development of real-time power reconfiguration and health-aware cooperative optimal control for multi-stack SOFC systems.

Abstract

Building façades represent a vast yet underexploited vertical surface for urban distributed photovoltaics (PV) beyond rooftops. Advances in radiation simulation, 3D urban modeling, and deep learning-based semantic segmentation have enabled batch acquisition of street?level façade data, including irradiance, window-to-wall ratio (WWR), material classification, and obstruction conditions. However, these multi-source heterogeneous data remain as numerical matrices and categorical labels
that planners cannot readily translate into building?specific retrofit decisions. This paper presents FaçadeRAG, a retrieval-augmented generation (RAG) framework that automatically converts structured façade data into traceable, rule-compliant photovoltaic planning recommendations. The framework integrates a domain-specific knowledge base spanning policy documents, technical guides, academic literature, and engineering case studies with a hybrid retrieval pipeline combining metadata filtering, BM25 keyword matching, semantic vector search, reciprocal rank fusion (RRF), and cross-encoder reranking. A constraint-guided generation module employing GPT-4o produces structured decision outputs with explicit risk identification, regulatory compliance checking, and evidence traceability. Evaluated on 15 façade segments across three streets in Shenzhen, China, with blind assessment by six domain experts, FaçadeRAG achieved 91.4% factual accuracy, 87.3% rule compliance, 95.2% decision completeness, and 93.8% evidence hit rate, significantly outperforming standalone LLM and naive RAG baselines while approaching expert-level quality at approximately 90 times the generation speed. These results demonstrate that RAG-LLM architectures can bridge the persistent gap between urban energy data acquisition and actionable planning decisions, offering a reusable paradigm for AI?driven building energy scenario assessment in high?density built environments.

Abstract

Near-zero-carbon communities require coordinated deployment of distributed energy technologies—such as building-integrated photovoltaics (BIPV), shared energy storage, and electric vehicle (EV) charging infrastructure—across multiple plots and building types. While technical design has advanced significantly, measures satisfying carbon thresholds may face stakeholder resistance due to divergent economic interests and risk perceptions. Conventional assessment methods—expert panels, Delphi surveys, and post?occupancy evaluations—are costly, static, and incapable of simulating interactive stakeholder dynamics. This study proposes a large language model (LLM)-based multi-agent simulation framework for rapid, quantitative, pre-implementation social acceptability assessment. The framework comprises four modules: issue extraction, stakeholder agent construction with heterogeneous utility weight vectors, three-round structured deliberation, and synthesis with design revision. Validated on a near-zero-carbon residential project in Shenzhen, China, six stakeholder types (24 agents)deliberated over four energy issues. BIPV cost allocation exhibited the highest conflict with a mean score of 4.8, while EV charger distribution reached near?consensus at 7.6. The three-round protocol drove convergence with inter-agent standard deviation reduced by 18%, without inflating scores. Agent?generated revision strategies improved aggregate acceptability by 27.4%, elevating all issues above the 7.0 threshold. Expert validation (n = 6) yielded Spearman ρ = 0.82 (p < 0.01) with identical issue rankings. This framework bridges the gap between technical carbon compliance assessment and socio-economic implementability analysis, providing planners with a low?cost, repeatable screening tool for community-scale decarbonization.

Abstract

Service restoration after fault isolation is critical to distribution network reliability, yet existing reconfiguration research stops at the final topology and overlooks the switching sequence that dispatch actually requires. This paper proposes a zero-shot LLM reasoning framework that generates an ordered, physically feasible switching sequence toward a target topology. Graph-theoretic topology logic and device constraints are embedded into prompts; a Chain-of-Thought paradigm decomposes the task into fault isolation, topology and outage analysis, service restoration, and load transfer; and a self-reflection mechanism validates feasibility and topological compliance. Validation on the IEEE 33-bus system confirms the framework’s effectiveness and interpretability.

Abstract

Sustaining rapid and stable reaction kinetics in metal hydride (MH) reactors during cyclic hydrogen absorption and desorption is essential for practical deployment, yet bed expansion and particle pulverization severely effect the cyclic reaction efficiency. Current mathematical models cannot fully capture their coupled effects, limiting long-term performance prediction and structural optimization of MH reactors. This study presents a novel mathematical model coupling bed expansion and particle pulverization with heat and mass transfer during hydrogen absorption. To reflect realistic bed evolution, the model incorporates dynamic changes in void fraction, particle size, thermal conductivity, permeability, and kinetic constants. Validated against experimental data, the model demonstrates high accuracy. Simulations indicate that bed expansion reduces the hydrogen absorption rate; neglecting this factor leads to a 6.7% overestimation of the average absorption rate. Additionally, a lower particle pulverization ratio (PR) reduces average particle size and accelerates reaction kinetics. Specifically, reaching 90% saturation takes only 1177 s at PR=0.238, representing a 40.6% time reduction relative to PR=0.714.

Abstract

Geometry of thermoelectric module (TEM) plays an essential role in designing TEM to achieve optimal efficiency. However, the effect of geometric parameters for TEM on output characteristics have not been thoroughly investigated when the prediction model considers variable thermoelectric properties, Thomson heat, air gap dissipation, and thermal contact resistance effects. Two important findings are as follow:(1)As the ratio of thermoelectric leg area to ceramic sheet area (φ=Apn/Acer) increases, the output efficiency shows parabolically increases rather than a non-linear increase when model considers air gap dissipation effect. The φ is strongly correlated with the height of the thermoelectric leg and increases with the height of the thermoelectric leg. (2) The maximum output power and efficiency occur at Ap/An < 1 when the Seebeck coefficient or conductivity of the n-type thermoelectric leg is much larger than those of the p-type leg. Meanwhile, when the thermal conductivity of the n-type thermoelectric leg is much larger than that of the p-type, the optimal Ap/An exceeds 1. The novel model and geometric analysis provide guidelines to improve and promote the performance of TEM.

Abstract

Against the backdrop of tightening water constraints and deepening interregional energy trade, energy production, processing, and consumption are continuously reshaping the spatial allocation of embodied water resources via commodity flows. Previous studies have mainly focused on aggregate virtual water accounting or provincial-scale transfers, yet limited attention has been paid to city-scale dominant corridors and core controlling nodes in energy-sector virtual water networks. Using the Yellow River Basin as a case study, this paper integrates a multi-regional input-output model, complex network analysis, and an attack model to construct a network-based analysis framework for the energy-sector virtual water transfer system across 61 prefecture-level cities in 2017. The network displays a pronounced hub-and-corridor structure, with outflows dominated by Lanzhou, inflows dominated by Lanzhou is the largest virtual water outflow city, with an export volume of 27.18 million m3, accounting for 46.01% of the total outflows, while Luoyang is the largest virtual water inflow city, receiving 21.27 million m3 and accounting for 36.00% of the total inflows. The Lanzhou-Luoyang corridor carries about 12.23 million m3 of virtual water, accounting for about 45% of Lanzhou’s total outflow and 57.5% of Luoyang’s total inflow, indicating that the strongest supply-demand linkage in the network is concentrated in a single core corridor. Centrality metrics further identify Luoyang as the leading topological hub, ranking first in degree centrality, betweenness centrality, closeness centrality, eigenvector centrality, and PageRank, while Qingdao, Weihai, Yantai, Chengdu, and Zhengzhou form important secondary nodes in different metrics. These findings indicate that the energy-sector virtual water network in the Yellow River Basin is characterized by hub concentration and dominant-corridor dependence. The framework can help identify key exporting cities, key importing cities, and dominant corridors, thereby supporting basin-level coordination of energy security and water ecological security.