To effectively suppress the transition of coal spontaneous combustion oxidation in goaf areas into the accelerated stage, a stable and high-heat-storage PCM was prepared using ammonium aluminum sulfate dodecahydrate and magnesium sulfate heptahydrate as raw materials. Through differential scanning calorimetry tests, solid-liquid transition tests, programmed heating tests, infrared spectroscopy, thermal insulation tests, and microscopic analysis, PCM thermal properties, the driving mechanism of solid-liquid transition, heat storage capacity, and coal oxidation inhibitory performance were studied. The test results show that the developed material has a phase transition temperature of 66.8℃, a latent heat of phase transition of 300.708 J/g, and a relatively low specific heat capacity during the heating process. The melting time exhibits an exponential decrease with increasing temperature, and a dense covering film forms after melting. When the material is added at a mass fraction of 15%, the inhibitory rate reaches 41%. Under a 500℃ heat source, with a material thickness of 30 mm, the effective thermal insulation time reaches 65 minutes.

In response to the frequent and high-impact accidents in key marine areas, a risk assessment model for accidents in such areas based on TAN network was established. To address the issue of partial sample bias in accident reporting, the boxplot method was employed to eliminate outliers and improve data quality. Considering the complexity and correlation of risk factors, a random forest algorithm was utilized to identify key risk factors and establish a risk evaluation index system for accidents in key marine areas. In addition, the performance of TAN network model was compared with six machine learning models for validation and analysis. The results demonstrate that TAN network achieves the highest accuracy of 93.02%. The findings indicate that ship speed, ship length, and pirate attacks are the primary factors contributing to risk events in key marine areas. Vessels aged between 11 and 20 years should be prioritized for maintenance and inspection. In addition, ships navigating in shallow key marine areas should operate with increased caution.

To address the safety risk issues of smart household appliance products, a systematic risk analysis was conducted by taking two sweeping robot-heater fire accidents in a certain location as the research subjects. An integrated method combining the FRAM and PLTS was adopted. FRAM was applied to analyze the accidents, through which nine core functional modules were identified, including navigation module, obstacle avoidance sensor module, and motion control module. Subsequently, PLTS was introduced to quantitatively determine the main characteristics of the input and output terminals of each functional module. Based on this quantitative analysis, four functional resonance modules that led to the accidents in the sweeping robot-heater were confirmed. The results show that the primary causes of the accidents include insufficient sensor accuracy and failure of users to implement the "power-off when leaving" operation. Additionally, the barrier system for risk prevention and control is identified, which consists of physical barriers, functional barriers, intangible barriers, and symbolic barriers.

In order to further prevent subway crowd stampede accidents and improve the safety management of metro stations, a metro crowd stampede resilience concept was proposed based on resilience theory. Centered on absorption capacity, resistance capacity, recovery capacity, and adaptation capacity, the developmental stages of metro crowd stampede resilience were analyzed, and a resilience evaluation index system was constructed by identifying core influencing factors related to crowd density. By integrating SD with MEE model, a metro crowd stampede resilience evaluation model was developed. This model was then applied to analyze the Beijing Xizhimen Metro Station. Results show that the model overcomes the limitation of traditional static evaluation models in depicting the dynamic evolution of the metro system's resilience capacity over time, enabling a quantitative assessment of dynamic resilience changes. The overall resilience level of Xizhimen Station is Grade II (relatively high resilience). However, due to delays in real-time monitoring and insufficient emergency response, it drops to Grade III (moderate resilience) during morning peak hours, with adaptability constrained by the low coverage rate of intelligent systems.

In order to explore the safety of interactions between pedestrian groups moving in different directions, a risk assessment method for pedestrian safety was proposed based on a Seq2Seq network integrated with an AM. By analyzing the interactions among pedestrians, local density was introduced as an additional input feature to better characterize dynamic interpersonal behaviors. The observed sequential features of pedestrians within a continuous time window were fed into an LSTM-based encoder-decoder architecture, and AM was employed to capture multidimensional critical information during motion, enabling the reconstruction of complex crowd movement patterns across various scenarios. Furthermore, pedestrian safety risks were quantified by introducing a crowd pressure metric, and the pressure value ranges corresponding to three typical movement states were calculated to enable risk stratification. The results show that the average displacement error (ADE) and final displacement error (FDE) in trajectory prediction are less than 0.3 m under both unidirectional and bidirectional flow conditions at different densities, indicating the model's high accuracy in trajectory prediction. Based on the risk variation observed in bidirectional flow scenarios, it is found that the probability of accidents increases significantly when pedestrians' velocities change abruptly under high-density conditions. Therefore, timely warnings of crowd states and proactive intervention measures are required to mitigate potential safety risks.

To address the limitations of relying on manual extraction of knowledge from complaint texts in automotive safety risk management, this study employed LLMs for automated risk discrimination. First, over 50 000 complaint texts covering eight major subsystems such as the engine were collected. A demonstration sampling method based on Bilingual and Crosslingual Embedding(BCEmbedding) model and community detection algorithm was proposed to construct a diverse and high-quality example knowledge base. Secondly, prompts were designed from perspectives such as scenarios, skills, and examples to develop agents capable of performing risk keyword extraction and expansion, as well as subsystem risk categorization. Finally, by analyzing reasoning knowledge from retrieved texts and utilizing chain of thought techniques, a risk level knowledge base for retrieved texts was established. A consensus-seeking multi-agent(MA) system based on LLMs was designed, resulting in a discriminative model for automotive safety risk levels. The results show that the model not only reduces labor costs but also achieves high accuracy and efficiency. It effectively supports risk term extraction, risk categorization, and risk level assessment in complaint incidents, thereby enhancing safety risk management.

To explore the time-varying performance of urban road networks under regular rainfall and establish a reasonable evaluation system for road network resilience, an LSTM model for the rainfall-induced traffic flow degradation in urban road segments was first constructed, and a network performance function with time-varying characteristics was second defined. Then, based on the resilience concept, a time-varying resilience calculation model for road networks was derived. Finally, the effect of rainfall on the time-varying characteristics of road segments and networks was investigated based on rainfall information and traffic data of a certain city and the Sioux Falls network. The results show that when the duration or amount of rainfall increases, the traffic flow degradation of road segments increases on weekdays. The road network resilience is a comprehensive reflection of the synergistic effect of various related road segments, and therefore its response to rainfall is slower than that of traffic flow degradation, and its value is generally not less than 0.9 under regular rainfall conditions. The road network resilience during low rainfall seasons is significantly stronger than that during high rainfall seasons, and the annual resilience of road networks is usually maintained at a higher level when not affected by rainfall disasters, and the difference between adjacent years is not significant.

In order to effectively respond to sudden mountain rainstorm disaster and explore the impact of emergency subjects' psychology on disaster evolution, a method combining game theory and BN was used to simulate the disaster's progression. First, historical disaster cases were analyzed, and consultations with relevant experts were conducted to identify four key scenario elements: emergency scenarios, disaster-forming environment, emergency measures, and human psychology. The Jaccard index was then employed to determine the relationships between these scenarios, while game theory methods were applied to optimize the BN node probabilities, which were initially estimated through case analysis statistics and triangular fuzzy number methods. This led to the construction of a BN-based disaster simulation model that incorporated the psychological factors of emergency subjects in the context of sudden mountain rainstorm disaster. Finally, the model was applied to analyze the "8·21" mountain flood and debris flow disaster in Jinyang, Sichuan, and comparative experiments were carried out. The results demonstrate that the disaster reasoning model is feasible and superior, and the neglect of psychological factors leads to an underestimation of disaster risks. Therefore, emergency management agencies should consider the influence of human psychology on early warnings, evacuation, and resource allocation when designing emergency strategies, in order to improve the effectiveness and adaptability of their measures.

To uncover the cross-system cascading propagation mechanism of urban waterlogging disaster chains and enhance blocking effectiveness, a coupled drainage-traffic network model was first constructed based on complex network theory, integrating high-precision topographic data, real-time traffic flow, and historical flooding records. A dynamic cascading-failure algorithm was then employed to quantify node centrality and link vulnerability, thereby dissecting the cross-system propagation path from "rainstorm-runoff-ponding-traffic paralysis". Finally, a three-level resilience blocking strategy—critical-node reinforcement, redundant-road-network optimization, and intelligent-response coordination—is proposed. The results show that pump-station expansion and redundant road design can effectively reduce waterlogged area and shorten traffic-interruption duration. And this enhances the system resilience and proactive immunity of critical urban infrastructures. It shifts the disaster-prevention paradigm from passive emergency response to active immunity.

To reval the disparities and shortcomings in safety and emergency management information disclosure, and to promote the strengthening of corporate safety responsibility and to improve of emergency management capability, textual information from 1 593 listed companies from 2008 to 2023 was collected from multiple sources, including CSR and annual reports of high-risk industries. Based on the three dimensions of reliability, relevance, and quantification, TF-IDF weighting method was applied to construct a 12-indicator evaluation framework for the quality of safety and emergency management information disclosure. The results show that corporate social responsibility reports provide more comprehensive safety emergency information and have become the primary disclosure channel. The overall quality of disclosure is relatively low. Only a few companies score in the upper range for disclosure quality, and the gaps between compaies are significant and continuously widenning. The disclosure content gradually focuses on the construction of emergency plans, emergency response and risk prevention and control strategies, but quantitative disclosure in areas such as safety investment and regulatory compliance remains insufficient. Significant differences exist among industries in terms of the number of companies disclosing information, focal points of disclosure and disclosure quality, with a lack of unified standards.