To address the issues of ambiguous multi-level，multi-link and multi-functional interaction relationships and the adverse coupling effects during operations in confined spaces，FRAM was introduced. Combined with ISM and AHP，the hierarchical structure and judgment methods were optimized and improved. By dividing the risk hierarchical structure，the impact of system function coupling variability was quantified，and the importance of functional units and hierarchical structures was calculated. Through the results of functional variability and coupling loss degree，the input-output phenotype of upstream and downstream functional changes were determined，and the coupling mechanism among system functional elements was clarified. The results show that by applying the risk prevention and control model for operations in confined spaces based on the improved ISM-FRAM-AHP，23 functional units and a 10-layer risk hierarchical functional network are obtained. The maximum values of functional variability and coupling loss degree are 4.36 (external environment F₂₃) and 0.808 4 (formulating operation plans F₂)，indicating a relatively high operation safety risk degree. The functional changes are mainly manifested in sequence，goal and control. Four effective functional barrier measures，physical，symbolic，functional and invisible，are proposed for 8 failure links.

Coastal chemical industrial parks have frequently experienced lightning strikes，which seriously threaten production safety. A risk assessment method based on HCL methodology was proposed to assess Natech (Natural-technological) accidents triggered by lightning strikes on storage tanks in these parks. Firstly，ESD and Fault Tree Analysis(FTA) were employed to qualitatively analyze the evolution paths of Natech accidents caused by lightning strikes on storage tanks，providing a visual foundation for blocking the accident transmission process. Secondly，BN were used to quantitatively calculate human error probabilities and assess the hybrid causal relationships of lightning-induced storage tank accidents. Finally，HCl methodology was applied to visually deconstruct the complexity and uncertainty of Natech accidents caused by lightning strikes on storage tanks. The research results indicate that decision-making error is the primary risk source in the human error model. Organizational atmosphere，psychological state，poor working environment and inadequate supervision are identified as the main factors leading to frequent human errors. The lack of effectiveness of lightning protection facilities is recognized as the trigger in the accident chain of lightning strikes on storage tanks. To reduce the severity of risk scenarios，emphasis needs to be placed on strengthening the control of full-surface fires and pool fires.

In order to address the challenges associated with characterizing the scenarios of stampede accidents and facilitating comprehension of these scenarios among decision-makers，a method for constructing and combining scenarios of large-scale event stampede accidents was proposed. Firstly，the scene elements of stampede accidents were extracted in large-scale events from the four factors that affect the formation of large-scale activities: people，venue，management，and environment，and a formal expression method for "state" and "trends" of large-scale activities research was established. Secondly，based on Markov model，a deduction description and calculation method for the transformation of situational "state-trends" was provided. Finally，an example analysis was conducted using Shanghai Bund accident. The findings of empirical analyses indicate that deductive results are largely aligned with the actual development process of the 2014 Shanghai stampede. This evidence substantiates the scientific rigour and efficacy of methodology proposed in the paper.

The S-FCN fire image detection method based on feature engineering was proposed to address the issues of high computational complexity and poor real-time performance of deep learning algorithms for fire image detection in complex backgrounds. Firstly，this method extracted color features from images in multiple color spaces and reduced the dimensionality of these features using mutual information. Secondly，the network structure of the deep learning model was simplified by using a single hidden layer of a fully connected network as its backbone. The color features in multiple color spaces can better represent fire smoke and flames，and reducing the dimensionality of color features in multiple color spaces effectively reduces the redundancy of input features. The single hidden layer fully connected network can significantly reduce the number of parameters during the model propagation process. Finally，this method was evaluated on a real and complex background fire image dataset. The experimental results show that the detection accuracy achieved by this method is 93.83%，and the real-time frame rate is 10 869 f/s. This method achieves high accuracy and high-speed fire image detection in complex scenes.

In order to identify high-risk stations in urban rail transit systems and improve network resilience and operational safety，a performance function model was constructed to evaluate the network resilience by selecting network efficiency，average shortest path length and maximum connection sub-graph as indicators，and an evaluation method for critical stations was proposed，considering topological structure and passenger flow distribution equilibrium. Taking Nanjing Metro as an example，three cascading failure modes，descending critical degree，descending betweenness centrality and random sequence，are adopted. The characteristics of resilience degradation under different cascading failures are simulated respectively in the unweighted network and the weighted network. The results show that the critical stations are similar on weekdays and weekends. The resilience performance decreases most rapidly in the early stages of failures in descending order of critical stations. Compared with topology networks，the resilience index of passenger flow-weighted networks decreases faster in the early stages of cascading failures. Strengthening the control of critical stations when cascading failures do not spread widely can help reduce the loss of network resilience.

To improve the city's resilience in responding to public safety risks，an integrated theoretical model of "spatial resilience" for urban public safety was proposed，and an urban public safety resilience evaluation index system was developed from the perspective of residents' perceptions. The entropy weight method was used to evaluate the public safety resilience level for Nanjing comprehensively. The results indicated that the overall score of Nanjing's public safety resilience was 6.851 5，and there was a structural imbalance in the resilience construction for the ternary space. Furthermore，social space safety resilience was the highest，followed by information space safety resilience，and physical space safety resilience was the lowest. The unbalanced development of urban public safety space resilience can be attributed to several factors including the lack of overall public safety resilience governance，insufficient investment in infrastructure and equipment maintenance，an imperfect social environmental risk monitoring system，and weak information security awareness and literacy among residents. Specific measures can be taken from the aspects of humanistic concepts，clear strategies，systematic thinking，and diversified cooperation.

To improve urban response capabilities in dealing with dynamic disasters，a MCMPOP was proposed for planning emergency vehicle paths in dynamic disaster environments. This model considered path safety as a multiplicative weight and vehicle path length and travel time as additive weights. Then，MCMPOP was addressed by improving the RSA. To verify the effectiveness of the improved RSA in solving the MCMPOP，510 simulation experiments were conducted comparing the computer time and solution quality of the Non-dominated Sorting Genetic Algorithm(NSGA)-Ⅱ and the improved RSA. Furthermore，"7·20" Zhengzhou rainstorm event was selected as a case study to validate the model. The results show that，compared to the NSGA-II，the improved RSA can find a complete set of Pareto optimal paths，effectively ensuring the optimality and computational efficiency of the algorithm. By using RSA to solve MCMPOP，it is possible to successfully select Pareto optimal paths with the shortest travel path lengths and the lowest time costs within the acceptable path safety range for emergency vehicles，providing more reliable routes for emergency vehicles and enhancing the urban emergency management capabilities.

In order to reduce the pedestrian safety problems caused by the lag of the emergency warning system in urban rail transit stations under large passenger flow conditions，the YOLOv5 algorithm was selected to predict passenger flow information. The artificial neural network (ANN) model was used to construct the urban rail transit emergency warning perception system. Firstly，the YOLOv5 algorithm was improved by optimizing the model training hyperparameters and prior frame parameters. Then，the emergency warning perception system was designed by selecting warning indicators，weight analysis and threshold definition. Finally，the self-organizing competitive network emergency warning model based on ANN was constructed by using Matlab software. The data collected by the optimized YOLOv5 algorithm were substituted into the emergency warning perception system through calculation，and the emergency warning perception system was verified by experiments. The results show that the optimized YOLOv5 algorithm can improve the accuracy of pedestrian target monitoring under large passenger flow conditions of urban rail transit by 7.04%. The judgment results obtained by substituting the pedestrian data collected by the optimized YOLOv5 algorithm into the constructed emergency warning perception system are consistent with the actual warning level，which proves the feasibility and effectiveness of the system and helps to improve the emergency warning level of urban rail transit.

Aiming at the problem that the regional environment of cultural relic buildings in a forest environment is complex，and it is difficult to accurately evaluate the regional firefighting and rescue ability，an evaluation method of firefighting and rescue ability of a single cultural relic building coupled with forest factors was constructed based on Analytic Hierarchy Process (AHP)，which covered 19 secondary indicators such as building materials，forest fire separation zones，and fire water sources. Based on single-unit assessment，a firefighting and rescue capacity assessment model based on the variable-rights method was established from the regional perspective for the centralized area of cultural relics buildings in the forest environment. A mixed-type variable-rights function dominated by penalties was introduced. The model was established by setting up the evaluation strategy，determining the penalties and incentives intervals，differentiating the protection level of cultural relics buildings，and putting forward the risk level division method. The firefighting and rescue capacity of the forest environment heritage building concentration area (Yuelu mountain scenic spot) was evaluated by combining the fixed-rights method and variable-rights method through field research and data collection. The results show that the number of individual heritage buildings rated as 'excellent，good，moderate，substandard，poor' accounted for '12%，20%，32%，20% and 16%'. For the centralized area，the fixed-rights method scored 75.27 (Level III，no need to take upgrading measures)，while the variable-rights method scored 69.97 (Level IV，need to take upgrading measures). Therefore，it is recommended that measures such as installing additional electrical fire warning systems and increasing forest fire breaks be taken to enhance firefighting and rescue capabilities.

To prevent occupational injuries of national mine rescuers，the epidemiological characteristics and risk factors of occupational injuries were investigated. Firstly，the injuries of the rescuers of the National Mine Emergency Rescue Team in Datong were investigated through field study，expert interviews，and questionnaire surveys. Then，a national mine rescuer injury causal model was proposed to determine the occupational injury risk factors. Finally，all risk factors were used in binary Logistic regression equation to determine the key risk factors. The results indicated that occupational injuries occurrence of the national mine rescuers was high (54.9%). Furthermore，training injuries (80.3%)，pulling and picking force devices (29.1%)，and weighted running (15.2%) were the most common training items for injuries. The waist (37.2%)，knee joints (11.2%)，hands (9.0%)，and ankle joints (7.6%) were the most vulnerable body segments. Work-related injuries (28.7%)，sprains (25.1%)，and strains (22.4%) were the most common injury types.The major internal risk factors affecting occupational injuries of national mine rescuers were training years，and the external factors were drinking alcohol，weekly training frequency，training with injuries，and lack of sleep. The major inducing events for injuries were insufficiency warm-up，recurrence of old injuries，and incorrect technical movements.