In order to reduce the number of accidents，casualties and enhance the ability to control the level of accident hazards，and to study the general characteristics and hidden patterns of heavy and large safety accidents in China's chemical production industry，this paper used statistical analysis to comprehensively analyze 41 heavy and large accidents that occurred from 2000 to 2023 in terms of time，region，production stage，type of accidents，causes of accidents and other elements. The results show that the number of accidents shows a fluctuating upward trend in 2007-2019，and July and August are the high incidence period of accident every year. The number of heavy chemical accidents in East China accounts for 52.2% of the whole country，of which 12 accidents occurred in Shandong Province，accounting for 57.1% of the total. The highest proportion of accidents occurred in the formal production stage of enterprises，accounting for 53.6% of the total. The main types of accidents is container explosion. The domino effect exists in accidents，with heavy domino accidents and large domino accidents accounting for 14.63% and 39.02% of the total accidents respectively. There are more accidents with domino effect in heavy and large accidents than those without domino effect，accounting for 53.65%，of which the casualties of accidents with large dominoes account for 40.1% of the total accidents and casualties. The most frequent cause of accident statistics is the illegal execution of production，accounting for 49.3% of the total，and the resulting accidents with domino effect account for 66.6% of the accidents. In response to the problems analyzed in the above accidents，this paper proposes some measures to improve the management system from three perspectives: the company，the equipment and the employees.

To enhance the intelligent system's understanding of individual driving behavior under human-machine interaction driving circumstances，an en-route driving style recognition method based on LDA model was proposed. The method explored vehicle trajectory information from multi-dimensions to quickly extract and identify latent driving style features of drivers. Firstly，the semantic understanding rules of driving behavior were established to discretize continuous trajectory data into semantic vocabularies of driving behavior，considering the scene perception layer，pattern layer，operation layer and vehicle status layer. Secondly，according to topic perplexity and consistency，habitual driving styles were classified into four categories: stable，conservative，moderate and aggressive. Finally，each driver's en-route driving style was identified as a probabilistic combination of the aforementioned driving styles. The results show that the proposed en-route driving style recognition method considers drivers' heterogeneity and explains the phenomenon of the same driver exhibiting different driving styles in varying driving environments. Additionally，this research improves the comprehensiveness and comprehensibility of en-route driving style recognition.

To effectively enhance the transportation capacity of compressed air foam and better address the fire hazards in converter stations，experimental research was conducted on pressure distribution using full-scale compressed air foam transport pipelines of 699，406 and 261 m as examples. Furthermore，the equivalent resistance length method was proposed to consider static pressure loss and local pressure loss，and a comprehensive equivalent resistance coefficient was introduced. An empirical pressure loss relationship suitable for engineering applications was developed. The results indicate that during the compressed air foam transport phase，the pipeline inlet and outlet pressures rapidly increase with time before stabilizing. The time required to reach a steady state is proportional to the length and complexity of the pipeline. After foam transport ceases，the pressure within the pipeline drops sharply. The complexity of the pipeline configuration leads to a nonlinear decrease in local pressure with increasing transport distance. Finally，an empirical relationship between the pressure loss and the equivalent resistance length has been obtained，which can be used to predict pressure drop variations for pipe diameters ranging from DN50 to DN200 and pipeline lengths up to 1 000 m.

To promote the all-around cultivation of undergraduate-master-doctoral integrated talents in safety science and engineering in universities，an effective collaborative education system was proposed by integrating curriculum ideological and politics，scientific research，innovation and entrepreneurship (dual innovation)，and industry education. Firstly，based on the current collaboration situation，the issues of the collaborative education system of safety discipline were deeply analyzed in terms of ideological education，science education，creative education，and industry education，and the connotation requirements of "a game of chess" were proposed. Then，a "four-dimensional integration" collaborative education system was developed through ideological and political guidance，research-driven，dual innovation cultivation，and industry-education orientation. Finally，based on the school's "Technology Mine" practical education brand，the practice of the "four-chain parallel" collaborative education system was performed in the case of safety discipline at China University of Mining and Technology-Beijing. The results showed that the collaboration education system for the safety discipline transferring from "four-dimensional integration" to "four-chain parallel" can meet the educational goals of the safety disciplines at different stages from undergraduate to master's to doctoral levels. It can also respond to the times for ideological and political，scientific research，industry，innovation，and entrepreneurship in safety science and engineering.

In order to facilitate the rapid formation of a disaster and accident trans-regional collaborative governance community in response to major emergencies，this study employed fsQCA. It investigated fourteen cases by selecting seven conditional variables across three dimensions: Incentives，mechanisms，and guarantees. These variables included value consensus，existential threats，administrative mobilization，collaborative linkage mechanisms，interest equilibrium mechanisms，legal guarantees，and digital intelligence technologies. The study explored the core influencing factors and the complex causal relationships in the formation of the disaster and accident trans-regional collaborative governance community. Configurational analysis identified three patterns of community formation: party-government-led，value-driven，and crisis-triggered，with each pattern corresponding to typical cases in the case database. The findings reveal that collaborative linkage mechanisms are a necessary condition for the formation of emergency management communities，while the other six variables cannot individually serve as necessary conditions. Optimizing the functionality of collaborative linkage mechanisms can effectively promote the development of trans-regional emergency management communities.

From the perspective of overall high-quality development and high-level security strategy，in order to realize the integration of new quality and productivity security development，research was carried out on the connotation，basic characteristics and capabilities of new quality and productivity security development. First，based on the triangle model，the connotation of safe development of new quality productivity was explained in combination with specific cases. Secondly，using the method of system engineering，the three-dimensional structural model was constructed，and the essential characteristics of the safe development of new quality productivity were analyzed. Finally，the capacities required for enhancing the safe development of new quality productivity were proposed. The research results indicate that the safe development of new quality productivity should focus on science and technology security，information data security，and risk management in new security fields，which are characterized by basic，overall，dynamic，complex and extensive features. The basic is reflected in the time dimension，the overall，dynamic and complex are manifested in the logic dimension，and the extensive features are embodied in the knowledge dimension. To promote safe development of new quality productivity，it is necessary to enhance strategic ability，leadership ability，collaboration ability，risk management ability，and theoretical innovation ability，corresponding to the three-dimensional structure model.

To address the problem of incomplete reflection of network structure characteristics and data distortion in static network models used in dynamic systems，an airport network structure analysis method based on a temporal complex network model was proposed by introducing time attributes. Network characteristic indicators such as nodal temporal degree，temporal centrality，temporal betweenness centrality，and temporal closeness centrality were analyzed. Then，the structural properties of CAN were investigated. Moreover，crucial airport nodes within the temporal network were identified and ranked，and the reasons for the heterogeneous node generation were analyzed. The results indicated that the temporal network structure of Chinese airports presented scale-free network characteristics. The average temporal distance values of CAN vary significantly with the maximum being 6.06 times the minimum. ZWWW (Urumqi Diwopu International Airport) and ZBHH (Hohhot Baita International Airport) represented relatively higher temporal betweenness centrality than other indicators，playing an important central role in the network. However，ZGSZ (Shenzhen Bao'an International Airport) had a higher temporal degree of centrality than other indicators.

In order to ensure the efficiency of response and disposal and reduce the damage caused by the hazardous gas leaks，it was necessary to quickly and accurately trace the location of the leak point as well as the source strength. In this article，a method of hazard location and source strength determination based on a complex algorithm was proposed. The basis of the algorithm was to compare the difference between the concentration of the monitored gas and the concentration calculated by the atmospheric diffusion model，and take the difference as the objective function，so that the parameter with the smallest objective function value was the optimal result of the source intensity and position. The results show that the complex algorithm can quickly and accurately obtain the location and source strength of the leakage source. Compared with the traditional simplex method，the compound algorithm has no restriction on the selection of initial value. Even if the selected initial value has a large deviation from the true value，the position and intensity of the source can be quickly obtained through the iteration of the complex algorithm，avoiding the shortcoming of the traditional simplex method，which has high requirements on the selection of initial value. A comparison of particle swarm optimization，genetic algorithm，simplex and complex algorithm is made in three aspects: traceability efficiency，traceability accuracy and traceability stability，which depicts that the complex algorithm is progressive. The complex algorithm can be applied to trace sources and determine source strength for continuous and instantaneous releases.

To avoid abnormal attitude problems such as serpentine and axis deviation during shield tunneling affecting construction safety，a large-diameter slurry shield attitude control method combining CatBoost algorithm and MOEAD. A shield posture prediction model was developed with 19 input parameters and 6 output parameters，and the CatBoost algorithm was used to develop a nonlinear mapping relationship between input and output parameters. The SHAP was used to analyze the effects of input parameters on shield posture. The CatBoost-MOEAD shield posture multi-objective optimization model was coupled with the multi-objective optimization algorithm. Then the proposed model performance was validated against the Wuhan Yangtze River large-diameter slurry shield tunnel project. The results showed that the CatBoost prediction model can efficiently predict the posture of large-diameter mud-water shields. The determination coefficients of the six shield posture objectives ranged from 0.931 to 0.974，the root-mean-square errors ranged from 0.030 to 0.880，and the errors ranged from 0.039 to 1.057. The thrust of the propulsion group has the most significant impact on shield attitude among the major construction parameters. The proposed CatBoost-MOEAD multi-objective optimization method for shield attitude had a great performance in optimization effect with a maximum value of 38.86%.

To address the challenge of extracting pulse signals in fault diagnosis of variable-load gearbox caused by redundant features，a pulse feature extraction method based on CAM was proposed. First，a CAM was designed，which consisted of two stages. In the first stage，a multilayer perceptron was used to simulate the channel dependencies and enhanced the important channel features related to faults. In the second stage，the convolutional layers were employed to learn signal segments related to faults. By recalibrating the features in two stages，the module focused on the critical pulse features. Next，based on CAM，this study proposed a CLDN method for extracting fault features in variable-load gearboxes. CLDN further improved the learning and representation of impulse signals by adaptively recalibrating the features at each layer. Finally，the extracted features were fed into a Softmax classifier to validate the feature extraction effect of the proposed method. The results show that CAM's accuracy is on average 3.8% higher than 4 attention mechanisms like Self-Attention，achieving accurate localization of impulse features. Compared with 7 diagnostic methods such as ResNet34，the accuracy of CLDN is 3.7% to 14.6% higher，which significantly enhances the extraction of fault features.