To address the problems of incomplete unimodal feature representation and low image-text fusion efficiency in construction safety hazard identification for hydropower projects, an intelligent image-text multimodal intelligent identification method was proposed. First, 12 categories of construction safety hazards were defined according to hydropower construction characteristics, and an image-text multimodal dataset was established. Second, bidirectional encoder representations from transformers (BERT) model and vision transformer (ViT) model were employed to extract hazard text and image features respectively. GFN was then introduced to dynamically adjust the contribution of image and text features and capture cross-modal correlated feature information, while a multi-layer perceptron was used to improve classification accuracy. Comparative experiments were conducted to verify the model's accuracy and reliability. The results show the method optimizes the contribution of multimodal features by enhancing identification stability. The multimodal hazard identification accuracy reaches 84.99%, representing an improvement of 1.73% over the text-based model and 12.24% over the image-based model.. The proposed approach outperforms existing benchmark models in hazard classification and improves the robustness of intelligent hazard identification.

To address the inability of existing battery management systems in acquiring internal temperature gradients, structural strain, and gas generation signals in lithium-ion cells, which result in inadequate safety monitoring and early-warning capability against thermal runaway, a systematic review of fiber optic sensors for lithium-ion battery safety monitoring and the challenges they face was provided. First, key safety-related monitoring parameters, including temperature, strain, gas evolution, and electrolyte state, were identified. Second, the principles and characteristics of representative sensing technologies, such as Fiber Bragg Gratings(FBG), Tilted Fiber Bragg Gratings(TFBG), Fiber Optic Evanescent Wave Sensing(FOEW), and Distributed Optic Fiber Sensing(DOFS), were reviewed. Third, the research progress in in-situ temperature and strain monitoring, electrolyte state evaluation, gas detection, and thermal runaway risk identification for each technology was summarized. Finally, major challenges in practical application, including integration compatibility, multi-parameter cross-sensitivity, long-term stability, and cost, were discussed. The survey reveals that fiber-optic sensing enables multi-point, millisecond-scale temperature monitoring with ±0.1 ℃ accuracy, strain mapping at ±0.1 με resolution, in-situ gas detection at 0.12% precision, and electrolyte refractive-index tracking down to 10^-3. Feeding these multi-parameter, in-situ, real-time data into advanced algorithms can significantly enhance early-warning capability for lithium-ion battery thermal runaway.

To accurately identify landslide deformation characteristics and triggering factors from monitoring data containing excessive noise, this study proposed a noise-reduction method for displacement monitoring data based on isotonic regression. It further established correlation rules linking deformation in different subzones to multi-level hydrometeorological conditions, incorporating time-lag effects. Using the Zhakoushi landslide in Fengjie County, Chongqing as a case study, displacement monitoring data before and after isotonic regression processing were comparatively analyzed to preliminarily investigate deformation patterns at different locations. The time delays between displacement at each monitoring station and rainfall and elevation of reservoir water level were calculated, enabling the extraction of association rules between deformation in these subzones and hydrometeorological factors, thus clarifying long-term deformation characteristics and its triggering mechanism of the landslide. The results demonstrate that the isotonic regression algorithm effectively removes non-physical noise while preserving intrinsic deformation information, considerably enhancing data quality. The landslide movements exhibit significant spatial heterogeneity, with the front part experiencing the most extensive deformation controlled jointly by reservoir drawdown and rainfall, followed by the rear part influenced by topography-enhanced rainfall recharge. The synergistic effect of rapid drawdown of reservoir water (>0.5 m/d) and intense rainfall (>30 mm/d), which generate an outward-directed seepage force and reduce matrix suction, is the primary triggering mechanism for the landslide.

To enhance the obstacle-crossing and terrain adaptability of mobile rescue robots in complex environments such as unstructured scenerios and mine disaster roadways, a variable-diameter wheel based on the waterbomb-origami principle was designed. Addressing the limitation of exisiting modeling approaches for waterbomb wheel structures, which rely on a single basic unit, a unified modeling theory was proposed that incorporates square, rectangular and parallelogram units as fundamental elements. By establishing multi-coordinate kinematic models for the wheel axle layer, wheel support layer, and wheel connection layer, and systematically deriving the corresponding constraint equations, a unified description and parametric analysis of the folding and deployment process of waterbomb wheels with different unit configurations was achieved. The variation trends and effective ranges of key state variables during folding and deployment were studied in detail, and deploy ability tests were conducted on prototype robot models fabricated via multi-material 3D printing. The results show that the kinematic modeling method is applicable to waterbomb wheel structures composed of different basic unit types. The maximum deployed diameter obtained from testing is 124.35 mm, with a 2.5% deviation from the design value, and the deployment ratio reaches 2.015. During testing, it was observed that the fold and deployment process of the actual structure closely matches the kinematic model, demonstrating the strong applicability and accuracy of the proposed modeling theory.

In order to effectively predict hydraulic support loads and evaluate the operational status of supports, a hydraulic support load prediction model based on MTAM-LSTM was proposed. The CEEMDAN algorithm was employed to decompose the load data of supports and extract intrinsic mode functions. Redundant components in the intrinsic mode functions were eliminated according to K-L divergence criterion, thereby forming the input sequence for load prediction. An MTAM was constructed to capture the variation characteristics of hydraulic support loads. Static attention generated attention weights for feature information of data, while dynamic attention optimized the focus on different sequence features. Residual learning was introduced to maintain the integrity of feature signals. LSTM networks were then utilized to establish deep dependencies between feature information and hydraulic support loads, enabling advanced prediction of support load data. Field data from the 402102 working face of a rockburst-prone coal mine in Shaanxi were used for empirical validation. RMSE, R², and MAE were used as evaluation metrics for comparison among different models. The results show that the RMSE and MAE of the MTAM-LSTM model are significantly lower than those of the comparison models, with RMSE reduced by 0.16-0.45 and MAE reduced by 0.16-0.45, while the coefficient of determination R² reaches 0.91 under different scenarios, thereby validating the prediction accuracy and generalization capability of MTAM-LSTM model.

To further investigate the effect of ultrasonic stimulation on gas diffusion in coal, this study utilized an ultrasonic stimulation test system for gas-bearing coal to analyze variations in gas diffusion under different ultrasonic conditions. Based on the Langmuir-like model, Quasi-first-order kinetic model, and gas dynamic diffusion model, the influence of ultrasonic stimulation on the kinetic characteristics of gas diffusion was systematically examined. Results demonstrate that increased ultrasonic power, elevated frequency, or prolonged stimulation time significantly enhance gas diffusion. Specifically, when ultrasonic power increases from 250 W to 1 000 W, gas diffusion rises from 0.689 mL/g to 0.981 mL/g, with the diffusion rate increasing from 5.92% to 10.69%. Similarly, gas diffusion escalates from 0.739 mL/g at 20 kHz to 1.074 mL/g at 40 kHz, elevating the diffusion rate from 6.36% to 9.65%. Extending stimulation time from 30 min to 120 min boosts gas diffusion from 0.833 mL/g to 1.100 mL/g, increasing the diffusion rate from 8.50% to 12.65%. The gas dynamic diffusion model exhibits the strongest fit for describing gas diffusion behavior under ultrasonic stimulation, followed by the Langmuir-like and Quasi-first-order kinetic models. Both the initial gas diffusion coefficient D₀ and its attenuation coefficient β demonstrate identical trends under ultrasonic stimulation, showing exponential positive correlations with ultrasonic power, frequency, and stimulation time. Ultrasonic stimulation enhances the kinetic characteristics of gas diffusion in coal by improving pore connectivity through mechanical vibration and pore-cleaning effects.

To address the frequent spontaneous combustion accidents of photovoltaic silicon sludge, a multi-experimental approach was employed to reveal its dual-stage exothermic characteristics and critical spontaneous combustion size. An isothermal microcalorimeter was used to analyze the heat release sources of silicon sludge under a low-temperature environment of 30 ℃, and to investigate the effects of pH and particle size on heat release characteristics. Simultaneous thermal analysis was applied to investigate the thermal behavior of silicon sludge during the programmed temperature rise process of 40-1 300 ℃, and the flynn-wall-ozawa method was adopted to determine the activation energy of the oxidation stage, revealing the high-temperature reaction mechanism. Based on metal basket self-heating tests and Frank-Kamenetskii theory, the critical spontaneous combustion temperature and critical size of silicon sludge under different ambient temperatures and sample states were calculated. Targeted safety control suggestions for the safe storage and transportation of photovoltaic silicon sludge were proposed. The results show that in the low-temperature stage (30 ℃), the heat release of silicon sludge is dominated by silicon-water and silicon-alkali reactions. The alkaline environment and small particle size can increase the maximum heat release power to 837.5 μW, significantly enhancing the heat accumulation risk. In the high-temperature stage (>405 ℃), the silicon-oxygen oxidation reaction becomes the main heat release source, and the oxidation activation energy decreases from 177 kJ/mol to 141 kJ/mol, with the reaction transitions from interfacial chemical control to diffusion control. The critical spontaneous combustion size of dried silicon sludge, alkali-containing silicon sludge, and silicon sludge with small particle size is significantly reduced, and for every 10 ℃ increase in ambient temperature, the critical spontaneous combustion size decreases multiplicatively. The minimum critical stacking size is 2.2 m at 60 ℃.

In order to enhance the emergency response capability of the firefighting and rescue teams and the overall efficiency of the forest and grassland fire prevention and control layout, an optimization method for the site selection of forest and grassland fire stations based on hybrid fire prevention emergency roads was proposed. By combining the eight-direction tilt point algorithm with digital elevation model data, a hybrid fire emergency road network was constructed to enhance the fire brigade's early prevention and emergency response capabilities. Subsequently, the location allocation model of the improved NSGA-II was adopted to optimize the site selection of the fire station, ensuring the rational allocation of resources and expanding the coverage. The results show that the coverage rate of the hybrid fire prevention emergency road in the overall area is 96.91%, and the coverage rate in the high-risk area is 93.51%, which improves the ability of the rescue team to deal with complex terrains. The optimized layout of the fire stations has a coefficient of variation of 0.26, ensuring the inspection and response capabilities of the teams. The overall demand satisfaction rate is 0.86, ensuring that the key areas are fully protected. The optimization model proposed in this study can provide a theoretical basis for the layout of forest and grassland fire prevention and control, improve the utilization rate of rescue resources, and promote the precise development of forest fire management.

To address the low efficiency and safety of personnel evacuation in fire scenarios amid China's accelerating urbanisation and the increasing number of densely populated venues, a single-story building fire evacuation model based on the A^* algorithm was developed. A grid-based model was introduced and combined with the PSO algorithm to comprehensively account for multiple complex factors influencing the evacuation process, including smoke concentration, high-temperature environments, hazardous gas dispersion, and personnel density. Based on the single-story building fire evacuation model, multivariate functional relationships were used to quantify the influence coefficients of various factors on personnel movement velocity, thereby achieving a precise description of the evacuation process. During numerical simulation experiments, the performance of the original and improved A^* algorithms was compared across scenarios with varying occupant numbers within the building. The results indicate that compared to the traditional A^* algorithm, the improved model reduces evacuation time by 20.9% and path length by 3.27%. It can effectively prevent evacuation paths from falling into local optima and avoid occupants entering dead ends.

To improve the efficiency of large-scale personnel evacuation and reduce casualties after toxic gas leakage accidents, a CFD model and wind direction statistical probability were adopted to simulate the spatiotemporal distribution of toxic gas concentrations. Combined with the facilities and personnel conditions of evacuation sites, a dynamic comprehensive risk measurement method for evacuation sites was proposed. First, based on the differences in dynamic comprehensive risks of evacuation sites, a multi-trip emergency evacuation bus scheduling model with splitable evacuation demand was constructed, with the objectives of minimizing evacuation time and total risk expectation. Then, the augmented weighted Chebyshev method was used to convert the multi-objective model into a single-objective model, and a genetic algorithm integrated with adaptive large neighborhood search was designed for solution, in which destruction and repair operators were adopted to improve the search capability of the algorithm. Finally, the rationality of the model and the effectiveness of the algorithm were verified through numerical example analysis.The results show that the emergency evacuation bus scheduling model with the objectives of minimizing evacuation time and total evacuation risk expectation can generate scheduling schemes that prioritize the evacuation of personnel from high-risk evacuation sites, thereby improving evacuation efficiency and reducing the total risk expectation. Compared with the expected total evacuation risk, evacuation time is more sensitive to changes in the latest time window of evacuation sites, and the number of required buses decreases with the extension of the evacuation time window. In comparison with the traditional genetic algorithm, the genetic algorithm integrated with adaptive large neighborhood search can reduce the objective functions by 4.46% and 2.44%, respectively.