In order to improve the interpretability of excavation deformation prediction，this study developed an interpretable machine-learning model aimed at predicting the deformation of excavation retaining walls. A comprehensive analysis was conducted to evaluate the influence of different feature variables on the prediction outcomes. Firstly，a large number of excavation support structure parameters were used as a dataset，and 80% of the dataset was used to build a prediction model for the maximum lateral deflection of the retaining wall using the XGBoost (eXtreme Gradient Boosting)model. Then，the model was tested based on the remaining 20% of the dataset，and the accuracy of the model was evaluated by four indicators，i.e.，the coefficient of determination，bias factor，mean absolute percentage error，and root mean square error. Finally，combined with the XGBoost model，the SHAP(SHapley Additive exPlanations) method was applied to complete the global explanation of the excavation feature variables，the partial analysis of individual samples，and the analysis of interaction effects of feature variables. The results show that the proposed method can provide both global and local explanations for the deformation prediction of excavation. At the global level，it not only provides the importance ranking of feature variables，but also gives the distribution of SHAP values. At the local level，the deformation prediction results of individual samples are decomposed into the base value and the contribution of each feature variable，which can quantify the impact of individual feature variables.

In order to solve the problems that machine learning exists in the hazard intelligent prediction field of tunnel water inrush，such as relatively simple models and imperfect prediction accuracy，a prediction model based on the stacking ensemble learning was proposed. Firstly，the tunnel water inrush disaster dataset was established by collecting 232 groups of water inrush disaster data from 95 tunnels，and the data was preprocessed. Then，3 base learners and 2 meta learners were selected to train 8 sets of stacking ensemble models in different combinations，and 6 sets of optimal ensemble models were selected. Finally，the optimal stacking ensemble model was selected by comparing and analyzing the prediction results of 6 groups of parameters optimized and stacking ensemble model with the grid search parameters and the 5-fold cross-validation hyperparameter optimization model. The results show that SVM(Support Vector Machine )+NB (Naive Bayes) + LR (Linear Regression) ensemble model is obtained after the optimal single model SVM is improved with the stacking ensemble learning method. Its accuracy，recall，and F₁ score are 0.94，0.91，and 0.92，respectively. The overall prediction effect is better than that of other compative models，and it can accurately predict the hazard level of tunnel water inrush.

In order to solve the problem of increasing difficulty of water injection in deep coal seam and poor inhibition effect of emission，based on the theoretical method of wetting modification of coal seam，wetting agent was used to act on gas-bearing coal body to inhibit gas desorption and migration. The fluorocarbon FS-3100 surfactant with strong wettability was used to test the influence of wettability on the gas-water migration process in coal through gas-containing coal desorption test and fracturing fluid displacement flow test，and to explore the change mechanism of coal wettability on gas-containing coal desorption efficiency and flow capacity. The results show that the surface tension of water is reduced to 17.9 mN /m after adding wetting agent in water，and the contact angle of coal water is only 3-3.5°. After the wetting agent was used to modify the coal body，the desorption rate of gas in the coal is significantly reduced. The gas desorption rates of Xinjing and Baode coals are reduced to 46.08% and 39.2%，respectively，which are 8.4% and 9.8% higher than that of water. After the wetting agent acts on the coal body，the displacement flow rate of the Xinjing coal sample increases from 8.59% to 14.10%，and the Baode coal sample increases from 10.65% to 16.67%，and the water injection capacity of the coal body is significantly enhanced. The wetting agent molecules are adsorbed on the surface of the coal matrix，which reduces the interfacial tension between coal and water and the surface energy of the coal body，so that the friction resistance of the water molecules flowing in the pores and fissures of the coal body is reduced，and the water molecules can infiltrate smaller-scale pores and produce stronger water lock effect. Finally，the gas desorption and migration adsorbed on the pores and coal surface are inhibited，and the gas plugging effect is formed.

To prevent the impact of external risks on cluster supply chain network，the propagation mechanism of external risks in cluster supply chain network was explored based on cascade failure theory. Feasible strategies to curb the spread of risks in cluster supply chain networks under external risk impacts were also explored through numerical simulation using Python. The results indicate that implementing risk tolerance enhancement strategies in important enterprises cannot prevent the eventual collapse of the network，but can effectively slow down the speed of risk propagation and reduce the impact of each step in the risk propagation process. However，the implementation of risk mitigation strategies by important enterprises can effectively prevent the spread of risks in the network and prevent the collapse of clustered supply chain networks. The implementation of risk tolerance enhancement strategies in the global supply chain can effectively prevent the spread of risks in the network and prevent network collapse. Although establishing a supply chain risk sharing mechanism cannot prevent cluster supply chain network paralysis under deliberate attacks，it can effectively reduce the number of failed nodes under random attacks and prevent cluster supply chain network paralysis.

To definite the diffusion characteristics of blasting fumes in downward drift filling mining stopes，taking the downhole filling mining method of Longshou Mine in Jinchuan as an example，numerical simulations and field tests were conducted. The spatial distribution of airflow in drifts and layered roads was studied，and the diffusion patterns of CO and NO₂ in drifts were analyzed. Furthermore，the effects of ventilation shaft locations and drift lengths on CO diffusion were explored，and the ventilation parameters of the Longshou Mine were determined. The results indicated that the airflow field in drifts and layered drifts can be divided into the inflow zone，neutral zone，and return zone. The airflow velocity in the drift. shows the S-shaped distribution，with higher velocity at the bottom，lower in the middle，and moderate at the top. In the vertical cross-section of the drift，the CO volume fraction continuously increases with height. While horizontally，it exhibits a "decrease-then-increase" pattern from the inner to outer side. At the drift waistline，the CO diffusion velocity shows a logarithmic decreasing trend with ventilation time.NO₂ is primarily concentrated below the midline of the drift，and its diffusion velocity is significantly faster than that of CO. The CO diffusion rate is negatively correlated with both the distance from the ventilation shaft to the drift entrance and drift length. When the distance between the ventilation shaft and the drift entrance is ≤40 m，and the drift length is ≤55 m，the CO and NO₂ concentrations in the natural ventilation blasting fumes are below the standard limits below after 30 minutes.

To address the current challenges of lacking unmanned detection systems amid frequent forest fires and inefficient personnel evacuation during uncontrolled fire scenarios，this article proposes a forest fire safety detection method based on collaborativeMUAVs and an optimized shelter location strategy. A dynamic forest fire spread model coupled with multiple influencing factors is developed on the NetLogo platform. MUAVscollaborative search mechanism，grounded in an improved ant colony algorithm，is enhanced by introducing attractive pheromones (guiding searches toward fire clusters) and repellent pheromones (avoiding redundant paths)，thereby optimizing the transfer probability of unmanned aerial vehicle (UAV) flight directions. Additionally，a flight model incorporating obstacle avoidance and water-carrying capacity-speed constraints was established. A dynamic evacuation simulation environment was constructed using geographic information system (GIS) data from Rhodes Island，Greece. Experimental results demonstrate that the improved ant colony algorithm reduces convergence time by 15% and 14% under 50% and 60% tree density scenarios，respectively，while search coverage increases by 35.02% and 32.16%. Furthermore，optimized shelter placement combined with the A^* algorithm-based evacuation strategy reduces the overall mortality rate by 2.525%.

To solve the issues that the bias，drift，gain，sticking and mutation fault modes of the current sensor in a battery pack are difficult to detect，recognize and evaluate，a comprehensive diagnosis strategy based on model fusion was proposed. A normal battery model with current as input and voltage as output (CIVO) was established. Based on the one-to-many relationship between the current sensor and batteries in the pack，the cumulative sum of the log-likelihood ratios of the residuals of the voltage of each cell was used as the detection index. A bias/drift fault model and a gain fault model with voltage as input and current as output (VICO) were established. Based on the residual variance of fault current，the model matching was performed on each fault mode. The quantitative evaluation of the bias，drift and gain modes were achieved by introducing a fault parameter to the fault model. The results show that based on CIVO，the five fault modes can be reliably detected. The sticking mode takes the shortest detection time and the drift mode requires the longest detection time，attributed to the slow-change characteristics of the fault current. Based on VICO，five fault modes can be accurately recognized. The quantitative evaluations of the bias，drift and gain modes are highly accurate，with the evaluation results of 0.396 2 A (experimental value 0.4 A)，1.641 7×10^-4 (experimental value 1.5×10^-4) and 0.201 6 (experimental value 0.2)，respectively.

In order to investigate the impact of drill pipe rotation on coal particles transport efficiency during negative pressure sampling，CFD and DEM were used to explore the effects of drill pipe rotational speed and coal particles mass flow rate on gas-solid flow characteristics. The results show that：the maximum axial airflow velocity inside the drill pipe decreases and stabilizes as the transport distance increases，while the maximum tangential velocity rapidly decays and vanishes. As rotational speed increases，the axial airflow velocity remains largely unchanged，but the tangential velocity significantly increases. With higher rotational speed，the spiral flow of coal particles becomes more pronounced，the length of the vortex region increases，and the length of the suspension region decreases. The number of coal particles entering the drill pipe decreases，and the distribution of particles above 5 m/s increases and becomes more dispersed. During rotation，the actual solid-to-gas ratio inside the drill pipe is lower than the set value，and coal particles transport efficiency increases and then decreases with the rise in rotational speed.

To improve the safety of air traffic operations，a delay level prediction method based on the combination of spatiotemporal association rule mining and deep learning was proposed. Firstly，the average flight delay time and delay rate were selected as airport delay metrics，and their spatial-temporal correlation characteristics were analyzed. Secondly，the airport delay levels were identified based on Fuzzy-C Means (FCM)clustering algorithm，and the spatiotemporal association rules of airport delay were mined based on (FP(Frequent Pattern)Growth) algorithm. Thirdly，sample data was constructed based on association rules and delay time series，which was put into LSTM model to predict the future airport delay levels. At the same time，attention mechanism was introduced into the prediction model to learn the influence of different rules on prediction. Finally，the actual US flight data were collected for example analysis. The results show that the average prediction accuracy of overall delay levels reaches 0.91 and the prediction accuracy of different periods is all larger than 80%. The connection weight of the attention layer network reflects the influence of each rule on the prediction，which can be used to explain the prediction results.

To address the oil-type gas emission hazard in coal-oil-gas coexisting mines，a quantitative risk assessment technology has been proposed. Firstly，the stability of the coal seam roof and floor strata was identified as a key parameter for evaluating oil-type gas gushing. The horizontal resistivity distribution uniformity was used to characterize strata stability. A direct current resistivity method was employed to investigate the resistivity distribution characteristics of the roof and floor strata，and a dynamic quantitative detection method for strata stability based on the resistivity variation coefficient was proposed. Additionally，comprehensive consideration of geological structures was integrated，supplemented by static parameters such as mining-induced damage depth，mechanical properties，permeability of the strata，and fault structures. Analytic Hierarchy Process (AHP) based on variable weight theory was employed to quantitatively calculate the weight of each factor relative to the evaluation indicators，thereby establishing a comprehensive quantitative evaluation method for oil-type gas gushing risks. Finally，the proposed method was applied to quantitatively assess the oil-type gas gushing risks in two typical areas of the Huangling mining area. The results align with field data from actual oil-type gas extraction boreholes，validating the method's reliability.