In order to improve the reliability of the logistics facility network of railway construction projects in complex environments，scenario reduction techniques were used to generate a minimum subset of disruption scenarios and their disruption probabilities to describe the disruption scenarios of transport channels. The polyhedral uncertainty sets were used to describe the uncertainty of logistics demand. To minimize the combined costs of transport，construction，operation and penalty costs，a two-stage stochastic and robust optimisation technique was applied to construct an uncertainty optimisation model for the location of material reserves bases. The model was solved based on a C&CG algorithm. The validity of the model and the algorithm was verified by taking a C railway construction project in a complex environment as an example. The results show that the cost variation coefficient of the model-acquired solutions is 4.3% of the traditional model in the random disruption scenario，and the cost fluctuation of the model-acquired solutions can be up to 38% of that of the traditional model in the extreme demand fluctuation. The two-stage uncertainty optimisation model given in this paper can effectively reduce the cost variation of the logistics facility network resulting from the disruption of transport channels and demand fluctuations.

In order to solve the problems that some operating conditions could not be automatically identified and the accuracy of abnormal operating condition recognition was low in the process of monitoring the production and operation of multi-product pipeline system，the intelligent operating condition recognition method was applied to construct a multi-product pipeline operating condition recognition model with real-time monitoring capability. First，logic rule discrimination methods and event logs in the multi-product pipeline system were used to supplement the data labels. Second，the data were segmented according to the start and end time of the operating conditions，and the subsequence of different operating conditions were extracted by using the sliding window. Third，the features of subsequence were extracted to construct the model for operating condition recognition of multi-product pipelines，and the recognition effects of six classification models，namely，random forest (RF)，adaptive boosting (AdaBoost)，support vector machine (SVM)，time series forest (TSF)，random interval spectral forest (RISF) and sequence learner (SEQL)，were compared and analyzed. Finally，a real multi-product pipeline was used as an example for model validation. The results show that the TSF model has the highest recognition accuracy for the four operating conditions of valve switching，valve internal leakage，pigging and sling pump，and is more suitable for the recognition of short-term operating conditions. In contrast，the recognition precision of the AdaBoost model has a higher probability of including the true value in the 95% confidence interval.

In order to enhance the efficacious operation of the air traffic control system，a quantitative model was established by focusing on the individual load of controllers. Tests were designed to collect pre-service and post-service data on various indicators from 16 area controllers in the front line. Sensitive variables were selected to describe individual loads based on changes in test data. A comprehensive assessment index system was established that included three dimensions: psychological perception load，physiological reaction load，and mental workload. The controller individual load index model was developed. The optimal weights of the individual load index were determined by the the entropy-critic combination weighting method. The quantitative model of the controller's individual workload was finally derived. Further K-Means clustering analysis was performed based on the controller's individual load composite index. There were evident discrepancies in the workload changes of the controllers due to different individual postures. The results indicate that the post-post individual workload changes of the controllers could be classified into three distinct groups. The first group，comprising 50% of the total number of controllers，exhibited the smallest post-post individual workload growth. The second group，accounting for 43.75% of the total number of controllers，exhibited a moderate post-post individual workload growth. The third group，comprising 6.25% of the total number of controllers，exhibited the largest post-post individual workload increase. These findings align with the instructor's ratings of controller competence.

In order to reduce the occurrence of safety accidents in coal mine production，from the perspective of emotion control，based on Valence-Arousal (V-A) emotional model，combined with emotional arousal methods and physiological measurement techniques，a cognitive experiment of miners' safety behavioral competence was conducted. Attention and decision-making time under different emotional states were measured. The regression analysis was used to investigate the continuous effects of degree-of-arousal on attention and risk preference under different emotional valence. The results show that in low degree-of-arousal and positive emotions，decreasing degree-of-arousal leads to weaker attention and more risk aversion in decision making in miners. In the high degree-of-arousal and positive emotions，with the increase of degree-of-arousal，the level of attention and risk aversion of miners in decision-making first increases and then decreases，until it is lower than neutral emotions. In low degree-of-arousal and negative emotions，decreasing degree-of-arousal would make miners pay less attention and have lower risk aversion in decision-making. In the high degree-of-arousal and negative emotions，an increase in degree-of-arousal increases and then decreases the attention and risk aversion in decision making，even until they are lower than the level of neutral emotions. By contrast，in the high degree-of-arousal range，increasing degree-of-arousal in positive emotion is more likely to reduce miners' safety behavioral competence to lower than the level of neutral emotional.

In order to study the reasons for the deterioration of the anchor performance caused by groundwater variation，a series of physical model tests were employed to explore the anchorage enhancement of the expanded anchor in the anti-floating process. Firstly，a single pull-out test was carried out through the indoor model test，and the change of bearing capacity of multi-bell-shaped expanded anchor under different buried depths during the pull-out process was obtained. Secondly，according to the ultimate uplift bearing capacity obtained from the single pull-out test，the cyclic test was carried out to explore the evolution mechanism of the bearing performance of the multi-bell-shaped expanded anchor under different cyclic amplitudes，cyclic times and cyclic frequencies. Finally，the image particle velocity method (PIV) was used to analyze the deformation mechanism of the surrounding soil，and the variation characteristics of the surrounding soil displacement under single pulling-out and cyclic loading were obtained. The test results show that the axial load-displacement curve of the bell-shaped expansion anchor can be roughly divided into three stages: elasticity，vibration and failure. With the increase of buried depth，the ultimate bearing capacity and soil displacement of the anchor increase. Under the action of cyclic loadings，the increase of the cyclic load ratio，the number of cycles and the cycle frequency will weaken the bearing capacity of the anchors.

To enhance the driving safety and achieve correct decision planning for autonomous vehicles，a safe driving trajectory prediction method based on EKF-GRU was proposed. By combining learning-based methods with physics-based approaches，the prediction accuracy was improved and the rationality of the predicted trajectories was enhanced. In the first step of this method，a prediction network was constructed based on GRU to predict the longitudinal acceleration and yaw angular velocity of vehicles by extracting historical trajectory features. In the second step，an EKF state estimator was built based on the nonlinear vehicle kinematics to generate the vehicle's future limited-time trajectory，incorporating the observations obtained previously. The trajectory prediction method was validated on the NGSIM I-80 and US-101 multi-vehicle trajectory datasets. Experimental results demonstrate that the final distance errors (FDE)，root mean square errors (RMSE)，and average distance errors (ADE) of the predicted trajectories generated by traditional physics-based methods are 6.48，7.69 and 3.03 meters，respectively. In contrast，trajectories predicted using EKF-GRU exhibit higher accuracy，and the corresponding values are 5.45，6.67 and 2.56 meters，respectively. This represents improvements of 15.90%，13.26% and 15.51%.

In order to address the increasingly serious cloud-native container security threats arising from large-scale cloud migration of systems，the ISM method merging IPFS，DEMATEL，and method were proposed to identify the key tactical factors influencing cloud-native container security threats and their hierarchical logical relationships from the security intruder perspective. The findings of this research are as follows: the centrality and causality of the persistence and privilege escalation tactical phases are high，positioning them at the core of the entire cloud-native security threat landscape. Security attacks during these two phases require high-priority attention. Threat attacks during the execution and persistence tactical phases constitute essential factors in cloud-native container security. The threats during the initial access，credential theft，and lateral movement tactical phases have the most direct impact on cloud-native container security. In comparison with traditional and triangular fuzzy sets improved DEMATEL-ISM，our proposed method has better performance in identifying container security-related critical factors.

In order to improve the effectiveness of hidden danger investigation，a reasoning model for hidden danger of explosive accidents of hazardous chemicals was proposed based on BN. First，according to the theory of combustion and explosion and expert experience，the main types of hidden dangers affecting the explosion of hazardous chemicals were determined，and BN model of explosion，leakage and ignition source was constructed. The prior and conditional probabilities of hidden dangers were determined by triangular fuzzy numbers. Then，Bayesian forward reasoning was used to calculate the probability of explosions of hazardous chemicals. Combined with reverse reasoning，the hidden dangers that were most likely to lead to explosion accidents were identified，and the formation of hidden risk factors was traced back. Finally，Bayes sensitivity analysis method was used to determine the key hazards affecting the explosion of hazardous chemicals and the correlation between the accident hazards，and it was verified by an explosion case of oil storage tank. The results show that the explosion probability of the storage tank calculated by causal reasoning is about 4.7%. Equipment overpressure and electric spark have the highest posterior probability，which is the most likely to lead to the explosion of the storage tank. The reason can be traced back to the failure to inspect the storage tank regularly or the failure to use explosion-proof equipment. Electrical spark and valve or flange damage are the key hidden dangers of oil storage tank explosion，and the accident hidden danger is related to the chain of "valve or flange damage → abnormal oil gusher → leakage exceeding limit → fire source (electrical spark) → explosion".

In order to examine the effects of typical optimization measures on the efficiency and safety of evacuating bottlenecks，evacuation tests incorporating pedestrian characteristics were conducted. The test encompassed 28 distinct cases，representing different combinations of optimization measures and pedestrian traits. Parameters such as evacuation time，speed，and local occupant density were measured across all conditions. Our findings reveal that the efficacy of bottleneck optimization measures is influenced by factors such as bottleneck width，the presence of luggage，and fixed evacuation directions. Practical implementation needs a tailored approach，integrating pedestrian characteristics and site-specific control strategies. Specifically，introducing a column in front of the bottleneck significantly benefits pedestrians without luggage，leading to a 15.30% reduction in density during bottleneck navigation with narrower widths，thereby enhancing safety，and concurrently improving evacuation efficiency by 13.18% in scenarios with wider bottleneck widths. Meanwhile，introducing a rail is preferable for pedestrians carrying luggage with wider bottleneck widths，especially when combined with a fixed evacuation direction，significantly enhancing evacuation efficiency by 21.90% while maintaining safety. Among the three bottleneck configuration alterations，incorporating a funnel-shaped passage preceding the bottleneck stands out as the most effective optimization measure，resulting in a notable 9.59% reduction in density，thereby enhancing safety，along with a simultaneous 9.14% decrease in evacuation times. It is noteworthy that the implementation of a straight channel or the combination of a straight channel and a funnel-shaped passage may yield negative impacts on both safety and efficiency.

In order to enhance the stability and safety of railway driving and effectively identify the influence of the dispatcher's fatigue state on the driving organization，a method for identifying the fatigue state of the dispatcher was proposed based on the characteristics of EEG signals. The fatigue state of the dispatcher was divided according to the working time period，and the high-speed rail scheduling simulation experiment was designed to collect EEG data. The three types of brainwave frequency-domain amplitudes of high-speed rail dispatching subjects were extracted as the characteristic value by wavelet series expansion and Fourier transform，and the classification results of fatigue state were verified by combining the operation characteristics and EEG signal characteristics of dispatchers. The ResNet18+SoftMax model and MobileNet V2+SoftMax model were built through the Python language environment. The input features were converted into a three-dimensional rectangular model based on deep learning. The weights were optimized and adjusted to obtain the optimal model，so as to judge the fatigue state of high-speed rail dispatchers. The research results show that the fatigue state recognition accuracy of the participants in the high-speed rail scheduling experiment by ResNet18+SoftMax and MobileNet V2+SoftMax two models is 92.78% and 99.17%，respectively，compared with support vector machines(SVM) model to improve the awake state and fatigue state recognition accuracy，and reduce the model computing time. Among them，the MobileNet V2+SoftMax model can better identify the fatigue state of the dispatcher. With the principle of MobileNet V2+SoftMax model as the core，the potential fatigue risk of high-speed rail dispatchers under long-term working conditions can be identified more quickly and accurately.