网络首发
Research on China's industrial carbon footprint flows and driving factors
Zhang Donghai;Under the context of the "Dual Carbon" strategic goals， the quantification and driving mechanism research of industrial carbon footprint is the core prerequisite for promoting the green and low-carbon transformation of the economy and society， and accurately formulating industrial emission reduction paths. Based on national non-competitive input-output tables， this paper systematically measures the carbon footprints of various industries and maps the flow of carbon footprints. Meanwhile， adopting the Structural Decomposition Analysis (SDA) method， it decomposes the changes in industrial carbon footprints from 2017 to 2023 into five effects: technological change， energy consumption intensity， carbon emission intensity， power generation structure and final demand.The research findings are as follows. First， inter-industry carbon transfer is a prominent feature in China. For resource processing industries， production-side carbon emissions account for 58.0%， markedly higher than the consumption-side share of 12.1%. By contrast， the construction industry registers 1.9% on the production side and 26.7% on the consumption side， while machinery and equipment manufacturing records 5.9% and 21.1% respectively. Second， the overall low-carbon transformation of national industries has achieved remarkable results. Energy consumption intensity effect serves as the primary driving force for emission reduction，cutting emissions by 1.93 billion tons. Carbon emission intensity， power generation mix and technological change also contribute to emission reductions (0.49 billion tons， 0.37 billion tons and 0.24 billion tons respectively). However， Final demand is the primary factor driving the growth of carbon footprints， pushing emissions up by 4.26 billion tons.Third， driving mechanisms vary distinctly across industries. The service industry offsets half of the carbon increment caused by scale expansion through improvements in energy consumption and carbon emission intensity. The construction industry basically hedges the carbon increment from demand expansion by virtue of technological progress as well as reductions in energy consumption and carbon intensity. Deepening industrial chain linkages contribute to rising carbon footprints in machinery and equipment manufacturing， resource processing， and the electric power and heat production and supply sectors.Agricultural carbon reduction mainly relies on technological progress and improvements in energy efficiency.The light industry achieves a substantial reduction in carbon emissions driven by improved energy efficiency， technological progress， and structural optimization. Technological upgrading serves as the dominant decarbonization driver for the mining industry. The findings of this study provide references for the scientific allocation of industrial carbon emission rights and responsibilities， as well as theoretical and empirical support for the formulation of differentiated and targeted industrial decarbonization policies.
Numerical study on the attenuation characteristics of explosion pressure waves by steel grating
GAO Chunyan;ZHU He;LIU Zhenyi;LIU Qiqi;LI Ranran;This study systematically investigates the attenuation characteristics of explosion pressure waves by steel grating structures in nuclear power plant indoor spaces using numerical simulation methods. The research focuses on the effects of grating angle， installation orientation， and incident overpressure on the propagation and attenuation behavior of explosion pressure waves. A three-dimensional transient compressible flow model was established in Fluent， and the SST k-ω turbulence model was adopted to simulate pressure wave propagation in a confined indoor space equipped with steel gratings. The incident pressure wave was simplified as a triangular pressure pulse with peak overpressures ranging from 5 to 20 kPa. Different grating angles (90°， 100°， 110°， and 122°) and installation orientations were considered to evaluate their influences on attenuation performance. The results indicate that the presence of steel gratings can significantly reduce indoor peak overpressure and modify the overpressure time-history characteristics， changing the pressure rise process from a sharp-rise pattern to a relatively gradual-rise pattern. The grating angle has a significant influence on attenuation performance. Smaller grating angles enhance pressure wave deflection， local reflection， and shear effects， thereby increasing energy dissipation during propagation. Specifically， gratings with angles of 90°， 100°， 110°， and 122° reduce the maximum overpressure by 51.66%， 45.32%， 35.06%， and 29.36%， respectively， while the minimum overpressure decreases by 21.08%， 18.04%， 13.63%， and 10.61%， respectively. In addition， installation orientation and incident overpressure also affect attenuation performance. Forward installation (long edge outward) provides better attenuation performance than reverse installation. As the incident overpressure increases， the relative attenuation efficiency gradually decreases， indicating that high-amplitude pressure waves weaken the attenuation capability of a single grating structure. The findings provide a theoretical basis for the anti-explosion design and optimization of steel grating layouts in nuclear power plant buffer spaces and similar industrial facilities， contributing to improved structural protection performance while maintaining ventilation and operational safety requirements.
Research on evolution prediction of high-rise building fire based on event logic graph
Wang Xingpeng;Ning Yongxing;This study aims to predict the evolution paths of high-rise building fires and enhance rapid response capabilities for such incidents. A total of 112 domestic high-rise building fire accident cases were collected， and a quantitative statistical analysis was conducted on the case texts. Each fire incident was treated as a unit， with the text segmented by punctuation marks. Event extraction was achieved using the part-of-speech tagging and dependency parsing modules of the HanLP platform， combined with self-designed predicate-centric multi-tuple extraction rule templates. To classify the fire events， four classification algorithms—Support Vector Machine (SVM)， Convolutional Neural Network (CNN)， Fully Connected Neural Network (FCNN)， and Random Forest (RF)—were compared. The extracted fire events were categorized into six groups: fire cause events， initial stage events， development stage events， spread stage events， decay stage events， and emergency rescue actions that span the entire fire evolution process. Precision， recall， and macro-F1 score were used as evaluation metrics to compare the classification performance of the four algorithms， allowing for the identification of the optimal algorithm suited for the fire event classification task in this study. The open-source large language model DeepSeek–R1–14B on the Cloud Studio platform was utilized to perform the event relation extraction task. The Bert–base–chinese model was applied to achieve vectorized representations of fire events， and within the framework of six predefined categories， the agglomerative hierarchical clustering algorithm was employed to generalize the fire events. A Neo4j graph database was used to construct the high-rise building fire event logic graph; betweenness centrality helped identify key nodes in the network， while the largest connected component index was introduced to further optimize key node identification. The cosine similarity algorithm was applied to predict the evolution paths of high-rise building fires， followed by a practical case analysis. The experimental results indicate that the CNN algorithm achieved the best event classification performance， with a macro-F1 score of 66%. The event generalization effect was optimal when the distance threshold was set at 8， reducing 2 059 events and 2 037 relations to 772 and 982 after generalization. A total of 17 key nodes were identified. This study effectively predicts the most probable evolution paths and fire causes during high-rise building fires and analyzes the differences between residential buildings and commercial office buildings. It is expected to provide a reliable theoretical reference and practical guidance for the field of fire emergency management.
Mine ventilation network calibration model with global convergence and its solution method
Wu Fengliang;Zhang Lei;Wang Ruitun;Cao Jun;Gao Jianan;To address the potential non-convergence issue that may arise during the calibration of a mine ventilation network model， we construct a nonlinear programming model with global convergence， which couples airway resistance and airflow rate， along with its corresponding solution method. The model employs the resistance correction term ΔR， the measured airflow rate correction term ΔQ， and the unknown airflow rate (Qᵤ) as its main decision variables. The objective function is fundamentally defined as minimizing the weighted sum of squares of ΔR. Equality constraints are formulated using loop air pressure balance and node flow balance equations， while inequality constraints are established using lower bounds for airway resistance. Optimal solutions are sought using the Sequential Quadratic Programming (SQP) method and the Interior Point OPTimizer (IPOPT). The square root smoothing technique is adopted to mitigate the adverse effects of second-order non-differentiability in the loop constraint equations on convergence. Corresponding software has been developed to calibrate a ventilation network consisting of 39 branches and 16 measured airflow rates. The calibration results demonstrate that the calibration function balances the ventilation network in a single step， achieving results that are difficult to attain manually. By incorporating a slack variable for the lower bound of resistance and a penalty term for the airflow rate correction into the objective function， we ensure that a feasible solution to the model always exists. When the penalty coefficient is set to (10⁵)， the penalty term becomes effective. Upon solving the model using both the SQP method and the IPOPT， the decision variables converge to the optimal solution from any initial guess. In this specific case， the SQP method exhibits superior convergence performance compared to the IPOPT. Furthermore， after applying square root smoothing to the airflow rates， the number of iterations required by the SQP method is reduced by 68%.
Hydrogen leakage and dispersion characteristics of a hydrogen–electric coupling system for PEM fuel cell applications
Luo Shunqin;Chen Aijun;Zhuo Yu;He Binbin;Wang Chunfei;Hydrogen leakage and hazardous accumulation in containerized proton exchange membrane (PEM) hydrogen–electric coupling systems were investigated using three-dimensional FLACS simulations. A model incorporating the equipment layout and ventilation grilles was established， with no ambient wind or forced ventilation. The porosity/distributed resistance approach and the standard k–ε turbulence model represented blockage effects and turbulent transport. An equivalent nozzle model described high-pressure releases， and grid sensitivity was assessed for a representative pipeline leakage case. Seven scenarios covered the gas–liquid separator hydrogen outlet pipe， the fuel cell system hydrogen inlet pipe， and the stack interior， representing low blockage， high blockage， and a small confined volume. Pipeline scenarios comprised micro-scale leakage， local rupture， and full-bore rupture， whereas the stack scenario involved a micro-scale release. Cloud evolution was evaluated over 10 s using hydrogen volume fraction， flammable gas cloud volume (FGC)， and equivalent stoichiometric cloud volume Q8. For local rupture in the low-blockage separator region， peak FGC and Q8 were 24.71 m³ and 14.89 m³， respectively. Full-bore rupture increased these values to 211.53 m³ and 73.36 m³， with a local volume fraction of 0.65 at 10 s. In the high-blockage fuel cell inlet region， local rupture produced peak FGC and Q8 values of 10.34 m³ and 6.42 m³， while the upper dead-zone volume fraction reached 0.365. A micro-scale release at a gauge pressure of 0.07 MPa inside the stack created an overall flammable environment within approximately 5 s. The equivalent stack geometry captured overall filling trends but introduced uncertainty in local transport associated with complex internal structures. Hazardous conditions could therefore develop before completion of the approximately 10 s detection–system response window. Three mechanisms were identified: rapid expansion driven by high source strength， localized accumulation under high blockage， and rapid filling of small confined volumes. Hazard assessment should jointly consider source strength， blockage， effective volume， and dispersion capacity.
Study on the impact of liquid CO2 cooling effect on the formation pattern of marker gases in coal spontaneous combustion
Liu Jie;Xia Fanfan;Zhang Zunguo;Qi Yun;To investigate the impact of liquid CO2 cooling on the spontaneous combustion characteristics of coal， a self-developed CO2 cooling simulation experimental system was utilized to treat preheated coal with both industrial CO2 and high-purity CO2. Programmed temperature experiments were conducted to analyze the generation patterns of two spontaneous combustion marker gases， carbon monoxide (CO) and ethylene (C2H4)， during the coal oxidation process following liquid CO2 cooling treatment. The results indicated that pre-oxidation and the freeze-thaw effect of liquid CO2 exerted both inhibitory and promotive effects on the generation of CO and C2H4， respectively. Furthermore， it was observed that at higher preheating temperatures， the volume fractions of CO and C2H4 generated during coal oxidation at the same temperature after liquid CO2 cooling were lower. As the temperature in the programmed tests increased， the volume fractions of CO and C2H4 generated from coal cooled by high-purity CO2—as well as the volume fraction of CO generated from coal cooled by industrial CO2—exhibited exponential growth. However， the growth trend of C2H4 volume fraction displayed distinct differences， which could be divided into three stages. Due to the oxidizing effect during the liquid CO2 cooling process， the low-temperature oxidation capability of industrial CO2-cooled coal was significantly lower than that of high-purity CO2-cooled coal. The initial appearance temperature of CO in industrial CO2-cooled coal was approximately 22 °C higher than that in high-purity CO2-cooled coal， and the volume fraction of CO generated at the same temperature was only 0.21 to 0.99 times that of high-purity liquid CO2-cooled coal. Industrial CO2 contained trace amounts of C2H4， which became enriched in coal samples through competitive adsorption during the cooling process. This resulted in the detection of trace amounts of C2H4 gas in industrial CO2-cooled coal at around 30 °C. The initial appearance temperature of C2H4 was found to be 92.9 °C to 98.2 °C lower than that of high-purity CO2-cooled coal， with the volume fraction of C2H4 being 3.06 to 18.39 times greater than that of high-purity CO2-cooled coal at the same temperature.
Knowledge-graph- and graph-neural-network-based ecological risk assessment and early warning for ski resorts
Xiong Jianhong;Kan Junchang;This study develops an integrated framework for ecological risk assessment and early warning in ski resorts， with emphasis on the bidirectional feedback between artificial snowmaking， environmental conditions， and operational safety. Meteorological， hydrological， soil， vegetation， remote-sensing， and operational observations are aligned on a common temporal and spatial reference. A domain ontology containing 12 entity classes and 40 relation types organizes these heterogeneous sources and explicitly represents reinforcing and damping feedback relations. The resulting knowledge graph contains 537，400 unique nodes and 1，284，630 edges after entity alignment and deduplication. Ecological vulnerability is initialized through the Exposure-Sensitivity-Adaptive Capacity framework， while its linear index is treated as an interpretable first-order approximation rather than an assertion of independent ecological effects. A serial forecasting pipeline applies Kalman filtering to estimate the current ecological state and uses a Long Short-Term Memory network to predict variable trajectories for the next one to four hours. Predicted states are written back to dynamic node attributes， and a relational graph convolutional network encodes risk propagation over relation-specific local subgraphs. Relation labels， polarity attributes， and relation-specific transformations distinguish reinforcing from damping messages， whereas the normalization constant only corrects neighborhood size. Bayesian updating and loopy belief propagation then revise event probabilities before the continuous risk score is mapped to four warning levels. Data are split by ski-resort-season groups， sliding windows are generated after the split， and all warning baselines use the same observation cutoff and one-to-four-hour forecast horizon. For knowledge extraction， the 30-shot prompt covers all 12 entity classes and yields an entity-recognition F1-score of 87.4%. For vulnerability assessment， the proposed method achieves a comprehensive F1-score of 87.6%， exceeding the Analytic Hierarchy Process–fuzzy comprehensive baseline by 17.0 percentage points. For risk warning， accuracy and F1-score reach 91.8% and 89.7%， respectively， with an average lead time of 4.5 hours. With the same one-to-four-hour forecast window provided to all baselines， this lead time is 1.5 hours longer than that of the Long Short-Term Memory baseline， and joint removal of the feedback mechanism reduces warning F1 by 12.0 percentage points. Within the observed ski resorts and validation settings， the framework improves interpretable risk-chain tracing and warning timeliness， although nonlinear ESAC alternatives， weight-perturbation analysis， and denser monitoring in ecological transition zones remain to be examined.
Research on landslide early warning in reservoir areas based on multi-source data and deep learning
Hu Bo;Wu Hongjiang;Na Jing;Zhao Feilong;Zhao Gen;Zhang Jingshuai;Frequent landslide disasters in reservoir areas pose severe threats to human lives， property， and the safety of engineering facilities， highlighting the urgent need to establish high-precision and timely monitoring and prediction systems. However， existing methodologies still suffer from significant limitations in large-scale spatial coverage， integration of heterogeneous data sources， and accurate characterization of complex deformation behaviors. To address these challenges， this study proposes a novel framework that synergistically integrates multi-source geospatial data with advanced deep learning techniques， taking Zigui County in the Three Gorges Reservoir region as the study area for empirical validation. First， Interferometric Synthetic Aperture Radar (InSAR) is employed for regional-scale wide-area deformation monitoring， enabling the preliminary identification of potential landslide-prone zones and the delineation of focused target areas for subsequent refined investigations. Subsequently， the Baijiabao landslide—a representative active landslide within the delineated area—is selected as the case study. Multi-temporal cumulative rainfall data and average reservoir water level records are integrated as exogenous triggering factors to comprehensively characterize the hydrological drivers of landslide movement. Using continuous displacement observations from the Global Navigation Satellite System (GNSS) as ground-truth references， a hybrid deep learning model termed CEEMD-TCN-BiGRU-Attention is developed. Specifically， Complete Ensemble Empirical Mode Decomposition (CEEMD) adaptively decomposes the non-stationary displacement series into a set of Intrinsic Mode Functions (IMFs) and a residual term， thereby mitigating mode mixing and enhancing signal stationarity. A Temporal Convolutional Network (TCN) then extracts local temporal features through dilated causal convolutions， while a Bidirectional Gated Recurrent Unit (BiGRU) captures long-range dependencies in both forward and backward temporal directions. Finally， an attention mechanism dynamically assigns higher weights to the most informative temporal segments， enabling high-fidelity capture of deformation magnitudes and evolutionary trends at critical observation points. Quantitative validation demonstrates that the proposed model achieves outstanding predictive performance across all key monitoring points， with a Mean Absolute Error (MAE) below 0.8 mm， a Root Mean Square Error (RMSE) of less than 1.6 mm， and a coefficient of determination (R2) exceeding 98.5%， confirming its superior capability in characterizing landslide deformation patterns. Overall， by fully exploiting the complementary strengths of multi-source remote sensing and deep learning， this study bridges the gap between regional "surface" monitoring and targeted "point" prediction， providing a practical technical solution for early warning and risk mitigation of reservoir-induced landslides.
Study on the influence of environmental temperature on the cycling aging behavior of lithium iron phosphate batteries for energy storage
Xue Feng;Zhao Junchao;Huang Xu;Fu Yangyang;Jiang Yong;Cheng Xudong;In this study， 100 Ah and 280 Ah prismatic aluminum-cased Lithium Iron Phosphate (LFP) batteries were selected as research subjects to systematically investigate their cycling aging behavior and electrochemical performance degradation across a wide temperature range of -20 ℃ to 60 ℃. A self-built experimental platform was utilized， which included a constant temperature and humidity test chamber， a battery cycler， an impedance analyzer， and an auxiliary temperature and voltage measurement device， ensuring the accuracy and stability of the tests. Specific testing methods included capacity calibration， Hybrid Pulse Power Characterization (HPPC) tests， Direct Current Internal Resistance (DCIR) measurement， and real-time surface temperature monitoring. These were employed to comprehensively analyze changes in key indicators such as battery capacity retention rate， voltage stability， internal resistance growth， and surface temperature during cycling. The results revealed that at 25 ℃， the batteries exhibited excellent cycling stability， with a slow capacity fading rate (capacity retention rate of 98% after 200 cycles) and stable internal resistance during long-term cycling. This performance was attributed to the favorable chemical stability of the olivine structure of the LFP cathode. At 60 ℃， however， capacity degradation accelerated significantly due to the decomposition of the electrolyte and severe aging of the electrode interface; the capacity retention rate dropped to 92.6% after 200 cycles， with noticeable increases in DCIR and polarization voltage， illustrating a "fast then slow" fading pattern. When temperatures fell below 0 ℃， the viscosity of the electrolyte increased sharply， significantly reducing the lithium-ion diffusion coefficient， impairing reaction kinetics， and causing a notable collapse in the voltage platform. At -10 ℃， capacity exhibited an exponential decay trend; after 45 cycles， the capacity retention rate was only 71%， and internal resistance had more than tripled compared to room temperature. At -20 ℃， the battery failed to charge and discharge normally after just 19 cycles due to severe nonlinear degradation. Notably， the 280 Ah battery demonstrated better low-temperature adaptability than the 100 Ah battery， primarily due to its greater self-heating effect during cycling， which effectively improved ion diffusion rates， reduced polarization losses， and mitigated the positive feedback effect of lithium deposition. Furthermore， most performance degradation observed after cycling above 0 ℃ could be reversed when the batteries were returned to room temperature， while long-term cycling below -10 ℃ led to irreversible capacity loss and structural damage to the electrodes. The findings provide significant theoretical and experimental insights for optimizing performance， predicting battery life， and designing thermal management systems for large-capacity LFP batteries used in energy storage applications in cold environments.
Risk identification and control for airport special vehicle operations based on text mining and Bow–tie model
Du Hongbing;Liu Sihui;Ni Shuai;Airport special vehicles perform essential support tasks for aircraft operations， and uncontrolled operational risks may threaten apron safety， potentially leading to serious incidents such as aircraft collisions and personnel injuries. To systematically identify and manage these risks， this study collected and analyzed 911 unsafe incident reports related to special vehicles from the Aviation Safety Reporting System (ASRS)， covering the period from 1999 to 2024. The research methodology employed text mining techniques using KH Coder software， which included comprehensive text preprocessing， topic modeling through Latent Dirichlet Allocation (LDA)， and co-occurrence network analysis based on Jaccard coefficients. This approach revealed intrinsic correlations and latent causal chains among various risk factors. The extracted causal chains were utilized as structured inputs to systematically construct the Bow–tie model， integrating Fault Tree Analysis (FTA) and Event Tree Analysis (ETA) to visually represent the complete risk evolution path from basic causes to potential consequences. The risk matrix method was applied to assess the likelihood of each identified risk factor based on frequency statistics derived from text mining， while its severity was evaluated according to Jaccard similarity with consequence themes， ultimately determining the risk level. Through text mining， ten key risk factors were identified， including inadequate communication， insufficient observation， distraction， vehicle speeding， unauthorized vehicle crossing， brake failure， towbar disconnection， insufficient lighting， unclear ground guidance lines， and insufficient driver training. Among these， three were classified as high-risk: brake failure， inadequate communication， and insufficient observation. The study established a comprehensive Bow–tie model for airport special vehicle operational risk， mapping the propagation path from basic events through intermediate events to the top event and its potential consequences. Additionally， it proposed targeted preventive barriers for different risk levels along with specific mitigating measures for potential consequences. This integrated text mining and Bow-tie model framework facilitates closed-loop management of vehicle safety operations， encompassing risk identification and effective control.