网络首发
Ship trajectory prediction model based on selective state space
Chen Wanzhi;Ao Xianyang;Li Xin;To address the limitations of current ship trajectory prediction models—including the neglect of variable correlations， mismatches between temporal dependencies and variable relationships， and inadequate consideration of complex marine environments leading to poor predictive performance—this study proposes a ship trajectory prediction model based on selective state space named MDMamba (Multi-Dimension Mamba). First， a spatiotemporal matching mechanism utilizing BallTree spatial indexing and Haversine distance is designed to integrate multi-source hydro-meteorological features， such as sea wind， waves， and currents obtained from publicly available sources， with trajectory data， thereby constructing a domain-specific dataset. Second， variables are treated as embedding units to preserve the independent characteristics of each navigation parameter， allowing the enhanced feature sequence to be transformed into multi-dimensional feature representations at both high and low resolutions. The low-resolution forward Mamba performs forward modeling along the variable dimension in the low-dimensional feature space， extracting abstract correlations and sequential information among navigation parameters while learning long-term trends and macroscopic patterns between variables. In contrast， the low-resolution reverse Mamba conducts reverse modeling along the variable dimension， capturing reverse dependencies among navigation parameters and compensating for the unidirectional limitations of the Mamba model. The high-resolution Mamba focuses on capturing fine-grained correlations among navigation parameters in the high-dimensional feature space， learning short-term dynamics and local dependencies between variables. Finally， a feedforward network independently learns the temporal associations of each navigation parameter sequence to achieve precise trajectory predictions. Using ship trajectory data from New York Harbor as the experimental subject， comparative experiments are conducted against Bi–LSTM (Bidirectional Long Short-Term Memory) networks， Bi–GRU (Bidirectional Gated Recurrent Unit)， the Mamba model， and the Transformer model. The results show that the MDMamba model reduces the Mean Squared Error (MSE) by 73.0%， 69.7%， 91.1%， and 46.1%， respectively， reflecting the smallest deviation between predicted and actual values. It also reduces the Root Mean Squared Error (RMSE) by 48.0%， 44.9%， 70.2%， and 26.7%， demonstrating greater prediction stability. Additionally， compared to the Transformer model， it reduces the Mean Absolute Error (MAE) by 19.5% and the average Haversine distance error by 144 meters， achieving higher predictive accuracy in geospatial terms.
Experimental study on fire spread characteristics of multi-layer cables with different ignition locations and layer spacings
Fan Chuangang;Du Linbo;Hu Junbo;Zhou Tiannian;This study aims to systematically investigate the fire spread characteristics of multi-layer cables， focusing on the influence of ignition location and layer spacing. The objective is to understand the complex three-dimensional fire propagation mechanisms and to assess how these parameters affect fire hazard. It can provide a basis for improved fire safety design and prevention strategies in cable installations. We constructed a full-scale experimental setup with a three-layer cable tray using three aluminum-core insulated power cables (YJLY 3×25). We designed twelve test conditions， varying the ignition location (center or end of the first， second， or third layer) and the layer spacing (0.2 m， 0.3 m， 0.4 m). High-speed and infrared cameras were used to record flame development. Thermocouple trees and arrays were used to measure temperature distributions. Key parameters analyzed included flame morphology evolution， average fire spread rate， and temperature profiles. For the condition with central ignition， the fire spread process exhibited four distinct stages: horizontal spread， vertical spread， flame merging of all three layers， and flame separation followed by a "V"-shaped pattern. End ignition resulted in smaller flames， a longer total burning time， and a unidirectional spread pattern. Peak temperatures (approaching 1 000 ℃) were primarily recorded above the second or third layer. The average fire spread rate was higher for central ignition and for multi-layer configurations. Increasing the layer spacing from 0.2 m to 0.4 m significantly reduced flame height and width and prevented upward vertical flame spread from the bottom layer. Ignition location and layer spacing are critical factors governing the fire behavior of multi-layer cables. Central ignition poses a higher risk due to faster spread and larger， merging flames. Increasing the vertical spacing between cable trays effectively mitigates fire risk by inhibiting vertical flame spread. The observed fire acceleration phase underscores the need for enhanced early-stage detection and suppression measures in cable fire safety management.
Research on lane-changing model of vehicles on rainy day expressway based on safety potential field theory
Zhang Huiling;Bai Nan;Wang Jingfan;Liu Tong;This study proposes an improved lane-changing model for expressways under rainy conditions， based on safety potential field theory. A total of 9 847 valid lane-changing events were extracted from the open-source HighD dataset， which includes data under sunny， light rain， moderate rain， and heavy rain conditions. Statistical significance analysis was conducted on key vehicle operating characteristics， including speed， acceleration， time headway， Stopping Distance Index (SDI)， and lane-changing duration. The results indicate no statistically significant differences in operating characteristics between sunny and light rain conditions; however， significant differences were observed among light rain， moderate rain， and heavy rain scenarios. Focusing on moderate and heavy rain， the model integrates critical factors such as the ego-vehicle’s instantaneous speed and acceleration， the relative kinematic states of surrounding vehicles， and real-time rainfall intensity. To quantitatively address the impact of rainfall， the model incorporates two key enhancements: a dynamic safety distance correction mechanism that introduces a rainfall intensity index along with specific adjustment factors for moderate and heavy rain， and a repulsion field model enhanced by a sensitivity coefficient to strengthen repulsive forces under adverse weather conditions. A unified artificial potential field model is established by integrating a weather adjustment coefficient. Model parameters were calibrated using a genetic algorithm with a population size of 40， a maximum of 50 generations， and a convergence threshold of 10⁻⁴. Simulation and validation results demonstrate that the proposed model achieves high predictive accuracy for vehicle lane-changing trajectories and acceleration profiles across varying rainfall intensities. Parameter sensitivity analysis indicates that the model's mean square error remains stable despite fluctuations in gravitational and repulsive field parameters， indicating strong robustness. Predicted trajectories closely align with actual trajectories from the dataset， and acceleration prediction errors are predominantly concentrated around zero. Specifically， the model attains a root mean square error as low as 0.045， under sunny and light rain conditions， while maintaining low error levels of 0.077 and 0.055 for moderate and heavy rain， respectively. Furthermore， the dynamic weather adjustment coefficient remains stable despite fluctuations in rainfall intensity， ensuring consistent model performance and reliable generalization across diverse rainy scenarios.
Three-dimensional facial fatigue characteristics of individuals with different chronotypes under sleep restriction
Li Jingqiang;Tang Yilan;Existing research on facial fatigue detection predominantly relies on two-dimensional image-based visual cues， leaving the systematic analysis of three-dimensional (3D) facial geometric structures largely unaddressed in this field. Furthermore， few studies have focused on the moderating effects of individual circadian rhythm variations on manifestations of fatigue. This experimental study induced acute fatigue through a controlled sleep restriction protocol. We successfully employed advanced photometric stereo vision technology to collect static 3D facial morphological data from sixty healthy participants with distinct chronotypes under both normal and sleep-restricted conditions. Focusing on three fatigue-susceptible regions—the tear trough， nasolabial fold， and oral commissure pouch—we constructed four independent micro-geometric 3D facial feature indicators for the first time: mean curvature (H)， Gaussian curvature (K)， mean included angle (A)， and divergence (D). The standard Karolinska Sleepiness Scale (KSS) was simultaneously utilized to validate the mathematical correlations between these objective geometric indicators and subjective drowsiness scores. The final experimental results clearly demonstrated that sleep restriction triggered distinct geometric shifts in 3D facial morphology. Under acute fatigue conditions， the micro-geometric parameters H， A， and D increased significantly， whereas K showed a significant decrease. Regionally， the nasolabial fold area exhibited the highest statistical sensitivity to morphological fatigue accumulation， while the oral commissure pouch displayed the lowest responsiveness. Notably， a highly significant interaction effect was identified between individual chronotype and fatigue state; the overall variation amplitude of 3D facial morphology in evening-type individuals was significantly greater than that observed in morning-type and intermediate-type individuals. This research establishes an objective， quantifiable， and rhythm-differentiated approach for assessing 3D geometric features to identify individual fatigue risk among populations with diverse circadian phenotypes. The final statistical significance value (p) is less than 0.01.
Risk identification of overtaking conflicts for passenger cars on mountainous roads based on machine learning
Ji Xiaofeng;Pang Xingren;Guo Yashi;Qin Wenwen;Lu Mengyuan;Li Jin;To investigate the conflict risk characteristics associated with passenger vehicles changing lanes to overtake on mountainous two-lane highways， both straight and curved sections of a typical mountain road were selected as study subjects. Initially， high-precision trajectory data for passenger vehicles were collected using Unmanned Aerial Vehicles (UAVs)， resulting in the extraction of 392 valid overtaking events. To comprehensively assess the overtaking conflict risk on mountainous roads， an improved Two-Dimensional Time to Collision (TTC2D) metric was employed， establishing a threshold of 2.01 seconds (the 15th percentile) to identify severe conflicts. Subsequently， twelve feature variables related to vehicle kinematics and vehicle type were selected， and passenger vehicle overtaking conflict risk identification models were constructed based on eleven machine learning algorithms， all rigorously evaluated through cross-validation. Finally， the SHapley Additive exPlanations (SHAP) algorithm was utilized to interpret the optimal models， revealing the intrinsic mechanisms and interaction effects of key features on overtaking risks. The results indicate: (1) Overtaking behaviors display distinct spatial variations; straight sections are characterized by higher speeds and larger gaps， presenting a “high-speed explicit” risk， while curved sections involve shorter gaps and greater acceleration fluctuations， exhibiting a “low-speed implicit” risk. (2) The conflict risk identification model for straight road sections outperforms that for curved road sections. Notably， the Categorical Boosting (CatBoost) model achieves the highest accuracy of 96.8% and an F-measure of 0.957 on straight sections， whereas the Light Gradient Boosting Machine (LightGBM) model attains an accuracy of 93.0% and an F-measure of 0.890 on curved sections. (3) SHAP analysis reveals that the headway distance and speed difference between the two vehicles are the most critical determinants of conflict risk. Risk severity exhibits a negative correlation with headway distance and a positive correlation with speed difference. Moreover， the type of overtaken vehicle inversely impacts risk severity， and a larger speed difference during the overtaking separation phase significantly accelerates the dissipation of risk.
Research on dynamic evacuation route planning methods for fire scenarios during renovation construction of existing buildings
Tian Wei;Li Xinyang;Zheng Fukang;Zhou Linhong;Hu Jiale;Gao Mingzhe;This paper presents an enhanced A* path planning algorithm integrated with visual perception technology to tackle the challenges of dynamic environment modeling and discrepancies between planned evacuation routes and actual on-site conditions during fire emergencies in the renovation phase of existing buildings. First， a vision-based algorithm developed using OpenCV identifies pre-arranged benchmark markers deployed at the construction site to compute the perspective transformation matrix from the camera view to a two-dimensional evacuation grid. Subsequently， the YOLOv8 model is utilized for real-time object detection on construction site images， enabling the identification of various obstacles and hindrances. The pixel coordinates of the detected objects are mapped onto the evacuation grid model through the perspective transformation matrix， facilitating structured storage of object categories and spatial location information while achieving real-time mapping from the dynamic construction scene to the evacuation grid. To enhance the adaptability of the traditional A* algorithm in dynamic risk environments， the heuristic function is improved by incorporating the influence of different renovation construction states on evacuee movement speed into the cost estimation process. The modified heuristic function employs the ratio of distance cost to dynamically adjusted evacuation speed as the search criterion， allowing for the generation of evacuation paths that are both safer and more time-efficient under changing renovation conditions. This improvement enables the path planning process to better reflect the actual evacuation characteristics of personnel in complex construction environments. Finally， comparative experiments and simulation analyses are conducted to verify the performance of the proposed algorithm. The results indicate that， compared with the path planned by the classical A* algorithm， the escape path generated by the improved algorithm saves 7.2% of time on average， demonstrating superior adaptability and effectiveness in dynamic environments. This approach effectively addresses the limitations of traditional methods in dynamic environment modeling and provides an innovative intelligent solution for fire emergency evacuations during the renovation of existing buildings.
Full scale experimental study on fire characteristics of electric vehicles under different disposal conditions
Zhu Xianli;Zhang Jiaqing;Guo Yi;Sun Tao;Shang Fengju;Huang Yubiao;This study systematically evaluates the effectiveness of various isolation and suppression measures for controlling thermal runaway fires in Electric Vehicles (EVs) under charging station conditions. A full-scale experimental platform was constructed to replicate vehicle fires induced by battery thermal runaway. Three representative test scenarios were designed: a baseline free-burn scenario without intervention， a Compressed Air Foam (CAF) suppression scenario， and a combined water mist–fire blanket intervention scenario. A full-scale pure battery electric passenger vehicle was utilized to ensure comparability under identical environmental and geometric conditions. To characterize fire evolution and suppression performance， multidimensional parameters were measured， including temperatures inside the vehicle， temperature fields above the vehicle， surrounding ambient temperature rise， and radiative heat flux at a distance of 1 meter from the vehicle. These measurements enabled a comprehensive assessment of fire spreading characteristics， thermal radiation hazards， and the temporal and spatial effectiveness of each intervention method. In the free-burn condition， thermal runaway rapidly propagated from the battery compartment to the front section of the vehicle， reaching a peak temperature of 801 °C and radiative heat fluxes exceeding 31 kW/m. Re-ignition phenomena were observed， indicating a significant hazard to adjacent vehicles and surrounding structures. In the CAF intervention scenario， the foam layer effectively provided cooling and surface insulation， reducing the peak temperature in the front compartment to approximately 363 °C. The radiative heat flux decreased rapidly to near-zero levels and remained stable throughout the later stages of the test， demonstrating strong capability in suppressing sustained combustion and mitigating secondary fire risks. For the combined water mist and fire blanket condition， fine water droplets achieved rapid early-stage cooling and significantly reduced thermal radiation， while the fire blanket provided temporary isolation by covering critical burning areas. However， during the prolonged heat release phase associated with ongoing battery reactions， secondary temperature and radiation peaks were observed， indicating limited effectiveness in suppressing continuous internal heat generation. Comparative analysis reveals distinct differences among the three approaches regarding suppression timing， control range， and stability. Based on these findings， a hierarchical， time-staged， multi-point coordinated intervention strategy is proposed to enhance EV fire isolation and suppression in charging station environments. The results provide experimental evidence and technical guidance for developing practical emergency response measures for EV thermal runaway incidents.
Spatiotemporal risk evolution of corroded pipelines considering climate change
QIN Guojin;JIN Dan;WANG Yihuan;To clarify the mechanisms by which climate change affects the integrity of buried pipelines， this study developed a comprehensive probabilistic assessment framework grounded in physical mechanisms. The framework integrates predictive data from the 6th Coupled Model Intercomparison Project Phase 6 (CMIP6)， a corrosion-driven leakage model based on soil chloride-ion-induced electrochemistry， and a Monte Carlo uncertainty-propagation analysis method. The study used the Delta downscaling method and the CN05.1 grid dataset to bias-correct temperature data from the CESM2-WACCM model in CMIP6. It optimized the probability distribution model for key environmental variables using the Akaike Information Criterion (AIC) and verified the model’s rationality with quantile-quantile plots and Kolmogorov-Smirnov tests. Based on the Shared Socioeconomic Pathways (SSPs) simulation of environmental degradation scenarios， five typical climate cities in China， namely Chengdu， Wuhan， Yingtan， Fuzhou， and Guangzhou， were selected to evaluate the spatiotemporal evolution of pipeline corrosion rate and failure probability under different radiation forcing levels from SSP1-2.6 to SSP5-8.5. The Sobol index method was used to quantify the sensitivity of pipeline failure probability to parameters. The results indicate significant regional heterogeneity in the risk of climate-driven pipeline degradation and failure. Under the SSP5-8.5 high radiation-forcing scenario in 2100， the corrosion rate of pipelines in Guangzhou exceeds 4 mm/a. In contrast， in Chengdu and Wuhan， it is less than 3 mm/a， with regional differences exceeding 1 mm/a. The corrosion rate in all cities increases monotonically over time. The probability of pipeline failure continues to increase with rising annual average temperature and radiation forcing， exhibiting a significant spatial distribution characterized by “high in the south and low in the north”. The increase in failure risk in southern cities， such as Guangzhou and Fuzhou， is much higher than in central and western cities. Temperature is the dominant factor affecting the probability of pipeline failure， followed by chloride ion concentration， and pH has the lowest sensitivity. These findings quantify the potential threats of long-term climate change to infrastructure， providing a theoretical basis for pipeline lifecycle risk design， monitoring， and planning， as well as the development of climate adaptation strategies.
Explosion characteristics of methane-ethylene mixtures and CO? suppression research
Dong Bingyan;Chen Weiye;He Junwen;Tong Le;This study investigates the influence of ethylene blending on methane explosion characteristics and the suppression mechanism of carbon dioxide (CO?). Experiments and simulations were conducted at ambient temperature and pressure (T=298 K， p0=0.1 MPa) using a standardized 20 L spherical explosion vessel， complemented by Chemkin software employing the USC Mech II detailed reaction mechanism (111 species， 784 reactions). The effects of the ethylene blending ratio (R， ranging from 0 to 1) and CO? addition (0 to 20% by volume) on the explosion characteristics of methane-ethylene-air premixed mixtures were systematically examined， with the equivalence ratio maintained at unity across all tests. Experimental results demonstrate that ethylene blending significantly enhances the explosion intensity and reaction rate of methane. As R increased from 0 to 1， the maximum explosion pressure (pmax) rose by 16.5%， the maximum rate of pressure rise [(dp/dt)max] increased by a factor of 2.3， and the flame propagation velocity accelerated from 2.37 m/s to 5.47 m/s. CO? exhibited a pronounced inhibitory effect on explosions; however， its suppression efficacy diminished at higher ethylene ratios. For instance， at R=0.4， the addition of 20% CO? resulted in a 50% reduction in pmax， a 91.9% decrease in (dp/dt)max， and a 73.8% reduction in laminar burning velocity. Chemical kinetics analysis revealed that ethylene modifies the reaction network by introducing pathways such as R254 (C2H4+OH=C2H3+H2O) and R16 (HO2+H=2OH)， which collectively enhance the concentration of OH radicals and the rate of the crucial chain-branching reaction R1 (H+O2=O+OH). Conversely， CO? participates in reaction R31 (CO2+H =CO+OH)， consuming H radicals and thereby reducing overall reaction intensity. This research elucidates the kinetic mechanisms through which ethylene and CO? respectively enhance and suppress explosions by regulating OH radical concentration and key reaction pathways. The findings provide a theoretical foundation and engineering guidance for risk assessment and explosion suppression design in industrial environments where methane and ethylene coexist.
Preparation of Bidens Pilosa-based adsorbent and its adsorption mechanism for PM10
Chen Li;Dai Zhirong;Xiong Xiaoming;Cheng Zhaoxia;Du Xiyuan;Wang Tao;A biomass adsorbent was developed from Bidens pilosa for capturing airborne PM10. The material was modified using potassium chloride in combination with neutral protease， and the preparation process was optimized through a quadratic regression orthogonal rotation-combination design， resulting in a reliable regression model and optimal treatment conditions. The influence of particle size， adsorbent dosage， temperature， relative humidity， initial PM10 concentration， and contact time on adsorption performance was systematically examined. Furthermore， the adsorption isotherm， kinetics， and thermodynamic behavior were characterized. The structure and surface properties of the samples were analyzed using Scanning Electron Microscopy (SEM)， elemental analysis， specific surface area measurement (BET) surface area measurement， and Fourier Transform Infrared Spectroscopy (FTIR) spectroscopy. Results indicated that with a neutral protease concentration of 0.2 mg/mL， a KCl concentration of 0.1 mol/L， and a KCl soaking time of 21.5 min， the maximum removal rate (Ymax) reached 18.05%. Optimal environmental conditions were identified as follows: particle size of 380 μm， adsorbent dosage of 2 g， temperature at 24 °C， relative humidity of 35%， initial PM10 concentration of 500 μg/m3， and an adsorption time of 50 minutes. The adsorption process was determined to be a mixed physicochemical mechanism dominated by chemisorption， involving multilayer binding on heterogeneous surface sites. This process was both spontaneous and endothermic in nature. The Freundlich isotherm and pseudo-second-order kinetic model provided better fits to the experimental data compared to the Langmuir and pseudo-first-order models， further confirming the multilayer and chemisorption-dominated mechanism. Thermodynamic parameters—ΔG < 0， ΔH > 0， and ΔS > 0—consistently supported the spontaneous and endothermic nature of the adsorption process. The surface of the modified Bidens pilosa was rich in hydroxyl (-OH) groups， greatly facilitating PM10 adsorption. The PM10 removal efficiency of the modified Bidens pilosa was significantly higher than that of other tested materials at the 1% significance level， ranking as follows: modified Bidens pilosa outperformed activated carbon， bamboo charcoal， diatomite， macroporous resin， and raw Bidens pilosa in descending order. Even after eight regeneration cycles， the modified Bidens pilosa retained strong adsorption capacity， indicating its excellent reusability and practical potential for sustained PM10 control.