In order to assess and monitor flight risks in real-time，clustering analysis was utilized to explore the abnormal patterns embedded in QAR data，and the influencing factors of abnormal flight patterns of civil aircraft were analyzed. The Euclidean distance was employed to characterize the similarity between samples of QAR parameters，establishing an abnormal flight pattern recognition model based on K-means to define the deviation degree of abnormal patterns. By considering the number of fatal accidents and the proportion of deaths in global commercial jet accidents，in conjunction with the deviation degree of abnormal patterns，the duration of abnormal patterns，flight phases，the likelihood of unexpected safety events，and the severity of consequences following unexpected safety events，a quantified assessment method for civil aviation flight risks based on QAR data was proposed. The feasibility of abnormal flight pattern recognition and risk quantification models for civil aircraft was validated through the practical QAR data of a certain airline. The results indicate that abnormal patterns are more prevalent during the cruising phase and critical moments at the transitions between flight phases. Significant differences are observed in the distribution of abnormal flight patterns and risks across different flights and flight phases. The average total risk value for flights is 166.94，with outliers exceeding 386.97. The abnormal flight risk during the takeoff roll phase is relatively low，with an average of 5.95，while the risk during the cruising phase is relatively high，with an average of 93.46.

In order to systematically review the research and development status of safety resilience in the field of civil aviation at home and abroad，and deepen the research on safety resilience in the field of civil aviation，firstly，the concept of aviation safety resilience was explored by reviewing policy documents，standards and research literature related to air transport systems in recent years. Then the research and application of safety resilience in civil aviation airport，air traffic control，flight operation and other fields were discussed. Finally，the existing problems were analyzed and corresponding suggestions and prospects are put forward. The results show that the safety management of civil aviation has formed a relatively perfect system，and the safety resilience covers all stages of safety management before，during and after，but it still cannot meet the needs of perfect connection and integration with the existing safety management system. Current studies focus more on the resilience of airports and route networks. In terms flight operation，more measures are taken to improve safety resilience based on the actual flight operation. The basic research on aviation resilience assessment is relatively lacking，and the research on aviation personnel resilience at operational level is far from enough. In the future，relevant research should be carried out around the individual operation resilience，enrich the basic research of resilience assessment，further deepen the research and form a relatively stable discipline system，pay attention to the connection between safety resilience and safety management system，and assess individual operation resilience from the perspective of safety.

In order to deeply analyze the structural characteristics and collaborative mechanisms of the emergency response cooperation network for secondary and tertiary earthquakes，the "12·18" Jishishan earthquake in 2023 was taken as a typical case，and the social network method was used to analyze the emergency response collaboration network systematically based on the network structure，organizational relationships and organizational functions. The results show that the density of the Jishishan earthquake emergency cooperation network is low，and the cooperation between organizations is not close enough. There are 8 cohesive subgroups，the units of the same level are more inclined to cluster，and the units with similar functional attributes are more likely to form cohesive subgroups. The main body of the two or three level response is the provincial-level department，which is mainly the cohesion subgroup of the provincial command and coordination and emergency rescue functions. The cohesive subgroup formed by units at the national level mainly plays a coordinating and supporting role. Central enterprises and state-owned enterprises played an important role in this emergency response. However，it is necessary to break the administrative barriers and establish a cooperative emergency collaborative mechanism between the local government and the state-owned enterprises.

To enhance the accuracy and reliability of geological earthquake disaster events predictions，a predictive model combining knowledge graph with GCN was proposed. Initially，the knowledge graph for geological earthquake disaster events was constructed，and the multi-source disaster-related information was consolidated into structured data. Then，the KGCN model was employed for deep learning of entities and relationships within the knowledge graph，uncovering potential association rules to forecast the evolution of disasters. Finally，the effectiveness of the model was validated through a set of geological earthquake disaster cases. The results show that the predictive model combing knowledge graphs with GCN exhibits excellent effectiveness in forecasting the evolution of geological earthquake disaster events，especially in dealing with complex multi-source data. The information can be efficiently integrated，and potential relationships can be accurately uncovered by the model. Excellent prediction accuracy is achieved in various aspects，including disaster levels，casualty levels，and disaster victim categories. Notably，the accuracy in predicting the disaster emergency response levels reaches 89.92%.

To prevent and treat the early hearing loss of firefighters caused by occupational noise sources，the occupational noise exposure detection for firefighters was conducted，and the degree of hearing loss in firefighters was evaluated. The ratio of hearing loss in firefighters was quantified，and the relationship between the hearing test results of firefighters and occupational noise exposure was analyzed. The results indicate that the intensity of the noise source for firefighters exceeds the exposure limit of noise levels in the workplace of 85 dB(A). Among the hearing screening results of 50 firefighters，the number of people who falisd the distortion product otoacoustic emission hearing screening in both ears is 21，accounting for 42.0%，and the pass rate is significantly lower than that of normal individuals. The number of people with left/right ear pure tone audiometry hearing thresholds ≥ 26 dB is the greatest at 6 kHz，accounting for 30.0% and 26.0% at n =15 and n =13，respectively，suggesting high-frequency hearing loss in firefighters after occupational noise exposure.

In order to improve the accuracy of emergency consequence severity assessment，clarify the correlation between the risk causes and consequence severity in urban traffic emergencies，the improved discrimination model of emergency consequence severity (IDM-ECS) was constructed and experimentally verified. First，based on the IFSA，the risk causes of emergencies were screened to obtain the important risk causes such as train fulfillment rate，punctuality rate，and daily network passenger volume and so on. Secondly，the improved hybrid restricted Boltzmann machine(HRBM) model was used to calculate the relationship between different risk causes and the consequence severity，and the discriminative relationship between risk causes and the consequence severity was obtained by comparing the probability values. Finally，the dataset of rail transit emergencies was used as an experimental sample for validation. The performance was compared with four models，including Generating Restricted Boltzmann Machines (GRBM)，Random Forest (RF)，Deep Forest (DF)，and Light Gradient Boosting Machine (LightGBM)，in terms of recall，precision，and F₁ value. The results show that train fulfillment rate，punctuality rate，daily network passenger volume，line 5 section full load rate，line 10 section full load rate，signal failure，and vehicle failure are the seven optimal risk causes. The IDM-ECS model has an average recall of 90.55%，precision of 91.89%，and F₁ value of 91.06%，all of which are better than those of the comparison models.

To effectively evaluate the performance of emergency logistics suppliers，a method for evaluating emergency logistics suppliers was proposed based on supplier evaluation，incorporating prospect theory and interval numbers. Firstly，based on the characteristics of emergency logistics，an evaluation index system for emergency logistics suppliers was proposed from six dimensions: rapid response capability，cost control capability，product quality，delivery service，internal and external conditions of the enterprise，and flexible demand. Then，interval numbers were introduced into the evaluation of emergency logistics suppliers，and an evaluation method based on prospect theory and interval numbers was proposed. The Jaccard similarity coefficient was used to define the similarity of interval number. The maximum sum of similarity with the remaining solutions was used to determine the reference point of the value function for the attribute evaluation value of the corresponding solution. The deviation maximization theory was used to construct a multi-attribute decision weight optimization model based on interval number similarity，from which attribute weights were obtained. Finally，the value function was normalized to expand the scheme discrimination. The prospect value of each scheme was calcuted based on the obtained weight function and value function，and the advantages and disadvantages of the scheme were ranked. The research results indicate that the difference in prospect values between the optimal and worst suppliers calculated using the evaluation method is 0.383 8，while the prospect value difference calculated using statistical inference principles is 0.085 6. The difference of 0.298 2 shows that the proposed evaluation method expands the differentiation between options，helping decision-makers achieve effective decisions.

To explore the "last-mile" problem of grassroots emergency response capacity, a MCDM based on grey theory was proposed to analyze the rural emergency response capacity under conventional and unconventional states based on the resilience theory. Firstly, from the perspective of resilience, the influencing factors obtained from literature review and field investigation were selected and optimized to construct a model of the influencing factors of rural emergency response capacity. Secondly, the MCDM model was used as a framework to analyze the causality, logical hierarchy, and characteristic state of the influencing factors. Finally, the key factors for the enhancement of the rural emergency response capacity and the resilience of rural villages were identified through the multi-criteria decision-making analysis. The results show that the centrality of leadership team structure is 2.95 and the driving force is 6, which is a tier 1 factor. The centrality of village grid management is 3.08 and the driving force is 6, which is a tier 2 factor. The centrality of normative document development is 2.7 and the driving force is 5, which is a tier 3 factor. The centrality of village network construction is 3.54 and the driving force is 9, which is a tier 3 factor. Leadership team structure, village grid management, normative document development and village network building constitute decision-making intersections, which are key catch-alls for the improvement of village emergency response capacity and resilience.

To quantify the transportation risks associated with biological samples using UAVs, this study first identified 32 risk factors across five dimensions-human, machine, environment, management, and hazard-based on national standards and relevant literature. A BN for risk assessment was constructed using Netica software, with prior probabilities determined through expert knowledge and fuzzy set quantitative analysis. The proposed risk assessment model was then used for bidirectional reasoning and scenario analysis. A case study of a UAV company in Shenzhen was presented to evaluate the transportation risks of biological samples and identify key influencing factors. The results indicate that the risk probability of biological sample transportation, as calculated through forward reasoning, is approximately 2.203×10^-5. The primary risk factors are related to hazardous materials, followed by equipment and facility-related issues. The core risk factors influencing biological sample transportation include the size, quantity and weight of hazardous material packages, the temperature control effectiveness of specialized cold chain logistics boxes, the integrity of emergency response plans, emergency handling capabilities, safety management and education, and the presence of obstacles.

In order to solve the problems of complex structure, large scale and difficulty in balancing detection accuracy and efficiency of the current forest fire detection algorithm based on deep learning, a lightweight forest fire detection algorithm based on YOLOv5s was proposed. Firstly, an optimized background difference technique was used to eliminate the interference of fire-like objects in the background image, thus reducing the time required for image analysis. Secondly, a group blending strategy was designed to optimize the conventional convolution, and an efficient channel attention (ECA) mechanism and depthwise separable convolution were incorporated into the C3 module of feature extraction, which enhanced the ability of image feature extraction and fusion and at the same time effectively reduces the number of model parameters. Then, a dynamic non-monotonic focusing mechanism was used to optimize the WIOU loss function, reducing the harmful gradients generated by low-quality samples. Finally, sufficient experimental comparisons between the proposed algorithm and other algorithms on the constructed forest fire dataset. The results show that the proposed algorithm shows good generalization in various scenarios, and the detection accuracy of the flame target can reach 86.1%, which is 2.7% higher than that of the standard YOLOv5s, and the detection speed is increased by 11.4%, which effectively reduces the fire false alarm rate and enhances the detection performance of the model.