The properties and risk characteristics of grey rhinos have been deeply investigated and used，which are new emerging concepts and metaphor theories in risk management areas. To prevent and mitigate risks and challenges nowadays，it is urgent to systematically investigate the grey rhino-related contexts. The risk classification system，evolutionary mechanism，and management strategies of grey rhinos were explored by integrating existing theoretical models and risk classification approaches. The results indicated that the grey rhino risk was attributed to objective environmental factors and subjective cognitive factors. Moreover，the grey rhino risk can be divided into hidden，stationary，and collisional gray rhinos based on properties，cognitive，state，time，spatial，and subject perspectives. The risk development trend of each grey rhino species was evaluated based on the mechanism of risk perception，risk distance，and risk framework influence factors. Furthermore，an evolutionary framework of "accumulation-fluctuation-mutation" within a complicated system was proposed to develop risk management strategies and improve cognitive abilities by identifying hidden gray rhinoceroses that were not easily perceived. The innovative idea is strengthened to resolve static grey rhinos of normalized coexistence，and agile mode is created to deal with sudden bursts of collision gray rhinos.

In order to monitor the risk during the navigation of MASS，the safety control structure of MASS was constructed based on System-theoretic Accident Model and Process (STAMP). STPA was used to define the losses/accidents and system-level hazards，identify unsafe control actions，analyze loss scenarios，and construct an accident model for system state transition. FTPN was used to model the process model，and a given MASS navigation situation was used to obtain the relevant fuzzy time functions and to project the situational evolution of FTPN. A new risk level expression was introduced，and a two-dimensional path diagram of system loss/accident was used to visualize the real-time system risk level and system unsafe states transition paths. The results show that at the current moment of the set navigation situation，no safe water depth input，no updated collision avoidance path，unsafe heading and speed，and grounding are the highest risk system unsafe states and correspond to the four highest risk transition paths. The study shows that the FTPN process model driven by STPA can comprehensively assess the real-time risk level of MASS navigation. Visualize real-time risk with a two-dimensional path diagram of real-time losses/accidents of the system，which can monitor the unsafe system states during MASS navigation and describe their transition paths.

To address the limitations of large language models in safety engineering，such as the corpus size，input processing capabilities and privacy concerns，ChatSOS，a Q&A system based on large language models，was developed. Based on 117 explosion incident reports from 2013 to 2023，a vector database to enhance the system's capability was constructed. ChatSOS integrated prompt engineering and external knowledge base to retrieve and analyze relevant data from the database. Compared to ChatGPT，ChatSOS integrated the external knowledge base，so that the big language model could retrieve the relevant corpus from the database according to the user's input information and make in-depth analysis. The results show that ChatSOS excels in in-depth professional problem analysis，autonomous task allocation，and providing detailed summaries and recommendations based on incident reports. By combining with the external knowledge database，the limitations of the large language model's professional corpus in safety engineering are overcome，which prevents performance degradation associated with fine-tuning on new datasets，broadens the application of large language models in this field，and paves the way for future advancements in automation and intelligent systems.

To solve the problems such as forward tilt deformation，serpentine shape，axis deviation and correction during shield tunneling，which affected the safety and efficiency of shield construction，a multi-objective optimization method of shield attitude combining CatBoost and NSGA-Ⅲ was proposed. Taking Guiyang Metro as the background，22 influencing factors were selected as input parameters，and the nonlinear mapping function relationship between input parameters and shield attitude was established by using CatBoost algorithm. The importance of input parameters was evaluated by random forest (RF) algorithm. A CatBoost-NSGA-Ⅲ multi-objective optimization model was established to minimize the absolute value of the shield attitude，and the applicability and effectiveness of the proposed method were verified by a case study. The results show that the prediction model obtained by using CatBoost algorithm to train engineering measured data has high accuracy，and the R² range of 5 shield attitude targets is 0.916-0.943. By using the CatBoost-NSGA-Ⅲ multi-objective optimization method，the attitude of the shield can be optimized significantly，and the average value of the overall improvement is 53.34%.

In order to explore the negativity and universality of security incidents，a reverse thinking paradigm based on Negative Systems Theory in safety science was first applied in conjunction with the Ecosystem Theory from systems science. This approach identified the characteristics of intelligent systems—openness，inclusivity，interactive coupling，and dynamic balance—and clarified the logical framework for intelligent system security analysis. Subsequently，the sources of security incident hazards were identified from the perspective of negative systems theory，and the mechanisms behind security incidents were explained. The concept of cohesive coupling was introduced，and an s-shaped framework model was proposed to analyze the stability of subsystems within intelligent systems and investigate their underlying technological support. Finally，using the concepts of producers，transmitters，consumers，and decomposers，the dynamic balance of information flow throughout the entire process in intelligent systems was analyzed，collectively shaping the security analysis framework under the negative systems theory perspective. The results indicated that the security analysis framework established under the negative systems theory perspective could effectively analyze the secure and stable operation as well as the dynamic balance of information flow throughout intelligent systems，ensuring the comprehensiveness and reliability of intelligent system security analysis.

In order to solve the problem of insufficient research on the non-uniform mixed explosion characteristics of hydrogen leakage in the tunnel，the numerical simulation software Fluent was used to establish a tunnel model with the length，width and height of 60，6.46 and 5.5 m，respectively. The non-uniform mixed explosion characteristics of hydrogen at the ignition time of 75 and 100 s and at ignition positions above and behind the vehicle were studied. The results show that the hydrogen leakage volume fraction field is not uniformly distributed in the tunnel scenario. After the explosion，the pressure wave reflects from the tunnel wall and propagates to the exit at both ends of the tunnel. In the process，the intensity of the pressure wave decreases continuously. The propagation characteristics of the blast wave formed by different ignition positions are obviously different. The influence of the explosion location on explosion overpressure is greater than hydrogen volume fraction. At the same time，the overpressure of ignition behind the vehicle is generally greater than that of ignition above the vehicle. The peak pressure at the position 2 m behind the vehicle can reach about 100 kPa. The rear ignition position within 2 m will cause serious injury or even death to the human body，and personnel within 4 m of the rear ignition position will suffer different degrees of injury.

Inattentional blindness is a human error related to selective attention or inattention. In order to reduce safety accidents caused by inattentional blindness in high-altitude work among construction workers，30 participants were selected for an experiment to analyze the importance of inattentional blindness in the failure of safety risk perception，as well as the personality trait-related factors that affect inattentional blindness and the failure of safety risk perception，and to establish practical safety management strategies. The results show that inattentional blindness accounts for 50% of the failure of safety risk perception. In addition，personality traits have a significant relationship with inattentional blindness and failure of safety risk perception，with extraversion，conscientiousness，and openness significantly correlated with inattentional blindness and failure of safety risk perception. Workers with low extraversion，high conscientiousness，and low openness have lower proportions of safety risk perception failure and inattentional blindness.

To solve the problems of single prediction state and insufficient prediction accuracy of the traditional coal spontaneous combustion prediction model，a prediction model based on RBF neural network optimized by SSA was proposed. Firstly，the temperature programmed test was used to analyze the variation characteristics of the index gas of coal samples with temperature. The coal spontaneous combustion process was divided into slow oxidation stage (80≤t_i<120 ℃)，accelerated oxidation stage (120≤t_i<160 ℃) and intense oxidation stage (t_i≥160 ℃) with coal temperature as the node. At the same time，the grey correlation degree between the index gas and coal temperature in each stage of coal spontaneous combustion was analyzed. Secondly，the performance of Particle Swarm Optimization (PSO)，Grey Wolf Optimization (GWO) and SSA algorithm was tested by different dimension test functions. Finally，the superiority of the RBF neural network optimized by SSA algorithm to the coal spontaneous combustion prediction model was verified by using six mining area data. The results show that the grey correlation coefficients of CO/ΔO₂，CO and C₂H₄ with coal temperature are the largest in the slow oxidation stage. The grey correlation coefficient between C₂H₄/C₂H₆，CO/ΔO₂，CO₂/CO and coal temperature is the largest in the accelerated oxidation stage. The test results of three different dimensional functions show that SSA has better global search ability，stability and faster convergence speed compared with PSO and GWO. When the number of neurons is 5 and the number of iterations is 300，the prediction accuracy of the SSA-RBF neural network prediction model for the slow and accelerated oxidation stages reaches 99% and 93% respectively.

In order to explore the influence of road section characteristics on path redundancy and the redundancy ability of road traffic networks under emergencies，based on the traditional method of path redundancy quantity in road networks，considering the efficiency difference and overlap degree between effective paths，a redundancy evaluation model based on path diversity and redundancy ability under emergencies was constructed. The influence of different costs on path diversity and the influence of emergencies on redundancy ability were analyzed，and the sensitivity analysis of redundancy ability was carried out. The results show that the path diversity in the road traffic network depends on the type of travel path cost and the difference between the travel cost of the road section. The closer the travel cost of the road section is，the higher the number of effective paths is，and the higher the path diversity is. In the emergency scenario，the redundancy of the road network depends on the importance of the failed road section. The more important the failed road section is，the lower the network redundancy is. The increase of travel cost tolerance within a certain range helps to improve the redundancy capacity of the road network during emergencies. Too high or too low travel cost tolerance will not significantly improve the redundancy capacity of the network.

To effectively evaluate the safety risk status and optimize safety management strategies in university laboratories，a dynamic safety assessment method was proposed based on a multi-layered safety control structure model. Firstly，a multi-layer safety control structure model was developed to identify the causal factors and control deficiencies of fire and explosion accidents in the university laboratory. Then，a safety dynamic feedback evaluation method was developed integrating an accident model based on system theory and Bayesian networks. Finally，the proposed evaluation model was used to determine the risk probability，analyze risk factors affecting safety incidents，and identify key risk factors in university laboratory safety. The results indicated that the proposed evaluation method based on the safety control structure model covered all risks identified by traditional methods and identified potential risks. Furthermore，the incorrect operation of hazardous chemicals，violation of operation procedures during the experiment，and the aging，short circuits，or defects in electrical lines were key risk factors affecting laboratory safety risks of universities.