To propose effective safety cognition strategies and reduce safety accident occurrence，it is urgent to investigate the safety cognition evolution process and characteristics of cavern constructors. Firstly，factors affecting safety cognition were identified based on a complex sociotechnical system and the 24Model，then a safety cognition index system of cavern constructors was proposed. Subsequently，the weight coefficient was calculated by the Super Decisions software. Then，system dynamic (SD) theory was used to determine the interrelationship between factors. SD numerical model was developed including organizational safety management and other three subsystems，and a cognitive correction mechanism based on safety investments was applied to the SD numerical model. Finally，numerical simulations and sensitive analysis were performed by the SD model based on actual engineering examples. The results indicated that the overall safety cognition of cavern constructors changed from a downward trend to an upward trend by increasing safety investments and exerting the role of cognitive correction mechanisms with project processes. Organizational safety management has the greatest influence on the safety cognition of carve constructors，which was consistent with the assumption of 24Model. Therefore，the scientific performance of the SD numerical model was validated.

In order to explore the mechanism of the influence of work underload on the risk perception ability of coal mine monitoring inspectors，boredom proneness and mind wandering were introduced as mediating variables to construct a chain mediation model of the four. A structured questionnaire was made with four scales: work underload，boredom proneness，mind wandering and risk perception. Quantitative studies were conducted on questionnaire samples from 263 coal mine monitoring inspectors，and the sample data were analyzed using Pearson correlation analysis and mediation effect tests. The results show that work underload has a direct negative effect on the risk perception ability of coal mine monitoring inspectors. Furthermore，work underload also affects risk perception through the independent mediating effects of boredom proneness and mind wandering，and the chain mediating effect of boredom proneness and mind wandering. Therefore，timely detection and intervention of work underload，boredom proneness and mind wandering of inspectors can effectively improve their risk perception ability and prevent accidents.

In order to address the inefficiencies in manually classifying and analyzing inspection records about civil aviation supervision，a dual-channel feature extraction short text classification model was proposed. The model combined data augmentation techniques and character-word vector fusion. The model aimed to tackle classification issues related to people，equipment and facilities，institutional procedures and institutional responsibilities in civil aviation supervised matters. In order to tackle the issue of class imbalance，data augmentation algorithms were employed to generate new samples by transforming the original texts，thereby balancing the sample sizes across different categories. The word vectors and character vectors were fused by combining them at the character level，resulting in character vectors that retain word-level features. These fused character vectors were then fed into TextCNN and BiLSTM for feature extraction at different dimensions. By extracting features from both local and global perspectives，this dual-channel approach aimed to capture comprehensive and effective information from the inspection records dataset in civil aviation regulatory matters. Experimental results on the civil aviation regulatory matter inspection record dataset demonstrate that the proposed model achieves an accuracy of 0.983 7 and an F₁ score of 0.983 6. Compared with some existing word embedding models and character embedding models，the accuracy is improved by 0.4%. Furthermore，when compared with commonly used single-channel models，the accuracy is increased by 3%，which validates the effectiveness and comprehensiveness of the features extracted by the dual-channel model.

In order to deeply analyze the influence of pedestrian panic on the efficiency of group evacuation under emergencies，this paper analyzed and constructed a quantitative model of pedestrian panic，which was composed of two parts: self-panic and panic propagation. First，the model considered factors such as detention time，local density，the distance between pedestrians and exits，the density of people around and the spread of pedestrian panic. Then it was introduced into the social force model for improvement and optimization，and verified the validity by comparing with the classical phenomenon. Finally，the simulation analysis was carried out using Anylogic software. The results show that the improved model can better reflect the evacuation movement characteristics of pedestrians in panic situations. A moderate panic，such as when the panic degree is less than 0.3，can accelerate the evacuation speed of the crowd. A high panic，such as when the panic value is greater than 0.3，can exacerbate the bottleneck phenomenon，and the so-called "fast is slow" phenomenon occurs，thus reducing the evacuation efficiency.

To effectively identify and assess different types of risk events in the risk zoo，and compare the difference of different risk metaphor events，based on theory of complex system thinking and vulnerability，risk metaphor events were identified and analyzed from 6 dimensions of risk certainty，universality，periodicity，stability，tolerance and reversibility. An assessment model of risk metaphor events was constructed based on the theory of the public safety triangle. The results show that through complex system thinking，we summarize and classify the general characteristics of risk metaphor events，and establish a multi-dimensional quantitative assessment model of risk metaphor events，which provides the application ideas and method logic for the control of risk metaphor events and ensure the scientific and reasonable establishment of risk metaphor events. Finally，the 4P (Prohibit，Suspend，Postpone，Prevent) strategy of risk prevention is proposed for the different cognitive characteristics of risk metaphor events，which provides a reference for organizations to carry out risk event control.

In order to cultivate a good safety culture in the operation stage of the nuclear power plant，the evaluation index system of safety culture in the operation stage of the nuclear power plant was constructed by analyzing the operation characteristics of the nuclear power plant and summarizing the existing evaluation index system and evaluation model，which was divided into four levels: values，behavior，system and environment，and subdivided into 13 secondary indicators and 61 tertiary indicators. On this basis，considering the non-independence and possible mutual influence between indicators，a comprehensive method based on DEMATEL and ANP was proposed to determine the weight of the index system. The results show that this method can get the evaluation index weight of nuclear safety culture by combining the survey data，and identify that the key to improving nuclear safety culture lies in the safety awareness of decision-makers and setting an example，which provides guidance for the cultivation of safety culture in the operation stage of nuclear power plants.

In order to effectively prevent and control the key safety risks in subway shield tunneling construction and accurately judge which risk coupling scenarios were significant ones leading to accidents，an improved N-K model was proposed to study the coupling effect of safety risks in subway shield tunneling. The key safety risks of subway shield tunneling were identified by comprehensively using literature research，accident case analysis and expert interview. Based on the N-K model，a new coupling analysis model of subway shield tunneling safety risk was proposed，and the applicability of the model was verified by a safety accident case. The results show that the list of key safety risk factors of subway shield tunneling is obtained，including 4 first-level risk factors (i.e.，personnel-type，equipment-type，environment-type and management-type safety risks) and 21 second-level risk factors. The safety risk value of subway shield tunneling construction increases with the increase of coupling factors. Safety risk coupling values with four factors are the highest，followed by the coupling values of three factors，and safety risk coupling values with two factors are the lowest. Coupling scenarios with workers' low safety awareness and machine failure participating are more susceptible to safety accidents.

In order to improve the economic benefits of coal mining enterprises and reasonable arrangement of various safety inputs，a model to solve the correlation between indicators was constructed. Firstly，the influencing factors of economic benefits based on safety investment were analyzed，and a set of scientific and reasonable structural index systems was constructed on this basis. Secondly，the entropy weight method was used to determine the weight of each index. Then，in view of the correlation between indicators，based on the safety input structure，the entropy weight-Choquet fuzzy integral model based on a 2-additive fuzzy measure was introduced to calculate the economic benefit evaluation value. Finally，a coal mine was selected for empirical analysis to verify the feasibility and effectiveness of the constructed model. The results show that the constructed model is more reliable than technique for order preference by similarity to an ideal solution(TOPSIS) method. This model can take into account the mutual influence and correlation between indicators and obtain more accurate and scientific evaluation results.

The purpose of this research is to clarify the research progress of the theory and technology of pedestrian abnormal behavior recognition in public places. Firstly，with the help of China National Knowledge Infrastructure (CNKI) and the Web of Science (WOS)，a broad definition and universal characteristics of abnormal pedestrian behavior in public places were given. The existing research results related to abnormal behaviors were divided into three categories: harmful behaviors，dissociable behaviors and violations. Then，from the perspective of data and technological foundations，the existing abnormal behavior recognition methods were divided into four categories: artificial design，human skeleton，Red Geen Blue(RGB) images and wearable sensors. Secondly，this study sorted out the abnormal behavior datasets of mainstream populations both domestically and internationally，and analyzed the performance of relevant algorithms on the datasets. Finally，the limitations of existing research methods in available datasets and data fusion detection were summarized，and future research directions and optimization suggestions were provided. The results indicate that these four types of abnormal behavior recognition methods have their own advantages and disadvantages. It is necessary to construct a diversified，well-defined and high-quality international benchmark dataset of abnormal behaviors among the crowd. Future research should focus on robust and accurate methods，models，and algorithms for identifying abnormal behaviors，explore multi-dimensional data fusion complementary detection methods，improve the application scenario consistency and adaptability of the theoretical results of abnormal behavior recognition，and eventually enhance the level of public place crowd safety governance.

A comprehensive database of MWSA was established to facilitate the data management to reduce the frequency and severity of accidents in China. From 2010 to 2022，a total of 278 records，supervised by the State Council Security Committee，were collected and stored in the MWSA database. The distribution of accident characteristics was explored for the dimensions of accident time，space，industry and type，which were examined in this study. Using a comprehensive quantitative index，a quantile regression model was developed to identify factors that significantly influenced accident severity. The results show that in terms of spatial and temporal distribution，the number of accidents and deaths from June to September is at the peak，and the number of accidents and deaths on sunny days (7:00-18:00) accounts for as much as two-thirds. There are fewer accident records in Beijing，Tianjin，Jiangsu，Zhejiang，Shanghai and Fujian. Explosions and vehicle injuries notably stand out as the primary accident types. At a significance level of 0.05，the accident severity is correlated with various factors，including accident type，date，season，sunlight，weather，average temperature，company staff size and company establishment time. Weather and average temperature emerge as pivotal factors influencing low-severity accidents. Moreover，enterprises with less than 100 employees are more prone to severe accidents.