In order to improve the safety condition of highways，26 320 highway traffic accident records in France from 2018 to 2022 were selected as the research object. Three representative algorithms were selected to impute missing values in the data，including the RF algorithm，the expectation-maximization (EM) algorithm，and the K-nearest neighbors (KNN) algorithm. The impact of different imputation algorithms on data stability was compared based on the changes in variable variance before and after imputation. The Apriori association rule algorithm was then applied to analyze the causes of highway accidents with different severity levels using the completed dataset. The results indicate that after missing value imputation，the RF algorithm demonstrates superior stability. Compared to the model trained on the original data，the accuracy is improved by 5.66%，the recall rate is increased by 9.22%，and the F1 score is enhanced by 9.91%. It is found that passenger vehicles are more likely to cause property damage accidents; motorcycles are prone to cause injury accidents on roads with lower speed limits and fatal accidents on roads with higher speed limits. The use of safety equipment is significantly related to the severity level of accidents.

In order to reduce the risk of boil-over accidents in storage tanks and develop efficient prevention and control strategies，two-dimensional transient numerical simulations of the boiling process of oil-water two-phase system in closed containers were carried out using Fluent software. The effects of different tank types，initial water layer thickness，oil-water ratio and other factors on the boiling process were analyzed through numerical simulation. The phase distribution，temperature field and hydrodynamic behavior were studied during the whole process of the combustion triggering to the end of the boil-over. The results show that for the same tank type，the higher the water content of the oil，the smaller the intensity of boiling overflow. When the tank boils over，the oil-water interface undergoes violent instability，and the two phases of the medium in the container are interspersed with each other，presenting a chaotic state. The liquid near the tank wall of boils upward first，with the peak temperature of the liquid phase up to 400 K. When the horizontal storage tank boils over，the overall temperature of the liquid phase is lower than that in the vertical storage tank. The onset of boiling overflow in the horizontal tank occurs earlier than in the vertical tank. Therefore，the boiling overflow in the vertical tank is more dangerous.

To address the issue of natural coal ignition in goaf under normal fault geological structures，the oxygen consumption rate and heat release intensity of coal samples were measured using a temperature-programmed oxidation device. Based on a porous media model of the goaf and the gas component transport equation，a numerical model for CO₂ injection via side pressure into the goaf influenced by normal faults was established. The numerical model was used to simulate the mechanism by which the variation in the distance between the working face and the fault affects the width of spontaneous combustion oxidation band in the goaf，and analyze gas migration characteristics under different CO₂ injection locations and flow rates. The results indicate that as the distance between the working face and the fault increases，the width of the oxidation band initially increases and then decreases，reaching a maximum width at 70 m from the fault. With the increase of CO₂ injection depth，the oxidation band width initially decreases and then increases again，reaching a minimum width when the CO₂ injection position is 40 m from the working face. Furthermore，with the increase of CO₂ injection volume，the width of the oxidation band width decreases following a negative exponential trend. When the CO₂ injection rate is 1000m³/h and CO₂ volume fraction at the working face is below 0.4% for safety，width of the oxidation band reaches its minimum.

To improve the safety of urban medical waste recycling and disposal，an optimization methodology for the medical waste transportation network was proposed，where the facility location，vehicle-routing and vehicle acquisition were simultaneously optimized. Firstly，according to the environmental transmission characteristic of medical waste viruses and uncertainty of urban emergency response time，BP neural network model was used to simulate the dynamic evolution process of risk，and a four-dimensional prediction model was designed. Secondly，introducing vehicle volume and capacity constraints，an optimization model，minimizing the total cost and risk，was developed by two-commodity flow formulation. The solution procedure was also developed by improving the NSGA-II algorithm based on the complexity of model. Finally，a case study in Shanghai and several tests were provided to demonstrate the workability. The computational results show that the new model and approach can provide multiple efficient plans within 672 seconds，and they are sensitive to some parameters. Compared to the traditional risk assessment，new model can provide a reduction of 3.47% and 13.04% in total cost and risk respectively. Using the risk prediction technique，a decrement of 7.41% in total risk can be achieved when comparing to current policy. New algorithm can reduce the CPU time by at least 49.44% and keep stable performance in solving problems of different scales while comparing to traditional multi-objective optimal methods.

In order to improve the safety of existing large-span steel grid structure stadiums and gymnasiums，from the three perspectives of foundation，superstructure and enclosure structure，24 factors influencing layer indexes were constructed，and the safety evaluation index system was weighted by game theory comprehensive subjective and objective weighting method. The importance grade of components was divided by load transfer path analysis，and the critical value of safety grade of different indexes was strictly defined，and the matter-element extension evaluation model was established. Based on the matter-element extension model，the data collection and model application of five existing large-span steel grid structure stadiums were carried out，and the stadium was taken as an example for empirical analysis. The results show that the safety grade of the stadium is grade II under the normal load (working conditions 1，2)，but the safety grade is grade III under the extreme load (working condition 3)，and the corrosion and deformation of the components need to be monitored. The agreement rate between the evaluation results of the five existing large-span steel grid structure stadiums model and the on-site inspection results by experts is 91.35%. The research confirms that the model can effectively quantify the structural safety of existing large-span steel grid structure stadiums under complex working conditions.

To reduce traffic accident risks，on-road experiments were conducted to investigate the differences in potential risk perception ability between skilled and unskilled drivers under two typical risk scenarios：dynamic motorcycle-following and parallel overtaking. A wearable eye tracker was employed to collect drivers' dynamic visual parameters，with their visual characteristics analyzed across different scenarios. The results demonstrate that skilled drivers exhibit significantly stronger risk perception abilities than unskilled drivers in both scenarios. Specifically，in the dynamic motorcycle-following scenario，skilled drivers show a higher probability of fixating on distant areas ahead，enabling better prediction of upcoming traffic conditions. In the parallel overtaking scenario，skilled drivers display shorter fixation durations，along with greater horizontal search breadth and vertical search depth. Moreover，the proportion of fixation time on rearview mirrors is significantly higher for skilled drivers compared to unskilled drivers，indicating superior visual search efficiency and enhanced rear traffic monitoring capability. Evaluations using the grey near-optimal comprehensive evaluation method reveal that skilled drivers achieved significantly higher scores in hazard perception ability.

In order to improve the emergency management capability for the purpose of urban underground comprehensive pipeline corridor construction，the general framework of pipeline corridor safety system was established by analyzing the regional emergency management system within the pipeline corridor，utilizing the theory of knowledge synergy，and combining the regional synergistic emergency management system with the scenarios of comprehensive pipeline corridors from the perspective of the application in comprehensive pipeline corridors. The framework was structured around the core of “risk-emergency-crisis”. The work breakdown structure-risk breakdown structure(WBS-RBS) method was adopted to complete the identification and grading of the risk of emergency management of the pipeline corridor. The probability of the risk of emergency management of the pipeline corridor was evaluated using analytic hierarchy process(AHP) method. Finally，the safeguards required for the establishment and operation of the multi-agency emergency knowledge management system of the integrated corridor were described across three dimensions of organization，technology，and supervision. The results show that the knowledge synergy theory is demonstrated to have a great advantage in building a regional collaborative emergency management system model. The safety system framework of the urban comprehensive pipeline corridor is designed to predict all kinds of risks in the structure，equipment and operation of the pipeline corridor. It is also capable of maintaining safety and stability during emergencies.

To reveal the distribution characteristics of ceiling radiant heat flux in full-scale tunnel fires，the numerical simulation research was carried out. Longitudinal ventilation velocities，heat release rates and effective fire heights were changed. The distribution of ceiling radiant heat flux in road tunnel vehicle fires，the maximum ceiling radiant heat flux and its position were analyzed. The influences of the longitudinal wind speed，the heat release rate，and the effective fire height on each parameter and their important degrees were revealed. Results show that under the natural ventilation，affected by the heat release rate and effective fire height，the radiant heat fluxes under the ceiling above the fire source surface show a uniform distribution，or the law that the middle radiant heat fluxes are low while the two sides are high. Under the longitudinal ventilation，the ceiling radiant heat flux increases and then decreases. The peak value of the heat flux occurs near the center of fire source and the downstream of fire source. At large effective fire source heights，the peak values of the maximum ceiling radiant heat flux are obtained under the condition of wind speed lower than the critical velocity. While reducing the effective fire height and increasing the heat release rate，The peak value of maximum ceiling radiant heat flux individually occurs both when the wind speed is less than and when it is greater than the critical wind speed. When the wind speed below the critical velocity，the maximum ceiling radiant heat flux presents right above the surface of the fire source，and it is weakly affected by the heat release rate，longitudinal ventilation and the effective fire height. When the wind speed above the critical velocity，the maximum ceiling radiant heat flux increases with the heat release rate，and it is less affected by the wind velocity. The research results can provide some reference basis and data supports for tunnel fire prevention and control and tunnel safety design.

In order to solve the index ambiguity problem in the risk assessment process of downward layered mining filling body stability，the filling body stability assessment model was constructed. Firstly，12 factors affecting the stability of the filling body，such as cohesion，exposed area，stress ratio，etc.，were selected as risk evaluation indicators to establish the evaluation index system. Secondly，the cloud model theory was introduced to calculate the cloud digital characteristics of each indicator，and the comprehensive weight was obtained by using G1-RF，and a comprehensive evaluation model of the stability of layered mining filler was constructed based on the G1-RF combined empowerment cloud model to determine the stability risk level of the mine filler. Finally，the evaluation method was applied to the stability analysis of the mine filler in the actual project. The results show that the comprehensive evaluation method can effectively solve the problems of indicator ambiguity and weak relevance in the process of risk assessment，more accurately and quickly evaluate the stability of the filling body，and realize the visualization of the risk level.

In order to effectively prevent construction safety accidents，a safety responsibility network model of the five responsible parties was developed based on complex network topology characteristics to reveal the nonlinear coupling mechanism of their safety responsibilities，supported by accident case studies. First，relevant regulations，standards，and literature were analyzed to extract safety responsibility checklists for each party，and a responsibility matrix was constructed. Subsequently，Neo4j software was utilized to visualize the network diagram illustrating the safety influence relationships among the five parties in construction projects. Finally，topological metrics were applied to explore the interdependencies between the parties across multiple dimensions. The results indicate that the construction client plays a central and dominant role in project safety，with particularly significant influence on the contractor. The contractor's impact is primarily concentrated within its organizational boundaries，and its responsibility implementation is susceptible to external influences from other parties. The survey and design units indirectly affect other parties through the construction client. The supervision unit maintains systemic balance and regulation through monitoring and feedback mechanisms. The identification of critical decision-making nodes and passive response factors provides network topology-based evidence for differentiated management of responsible parties.