To improve the short-term prediction accuracy of photovoltaic power generation models with multiple input features, a photovoltaic power prediction ensemble model LGGWO-TCN-MHSA based on optimizing TCN hyperparameters was proposed. The model integrated the levy gold grey wolf optimization (LGGWO), temporal convolutional network (TCN), and multi-head self-attention mechanism (MHSA). First, the Spearman correlation coefficient method extracted the main features that significantly affect photovoltaic power, which were then fed into the TCN prediction model. Then, the proposed multi-strategy LGGWO was applied to the TCN for hyperparameter optimization, which improved the model's prediction performance. Finally, the predicted values were input into the multi-head self-attention model to further boost prediction accuracy. The experiment was verified using original Australian photovoltaic data. By comparing with six groups of models including convolutional neural networks (CNN) and long short-term memory neural networks (LSTM), the mean absolute error (MAE) and root mean square error (RMSE) of the proposed model on the test data set were reduced by 2.03%~82.0% and 10.5%~80.1%, respectively. The results show that the proposed method has high prediction accuracy and good stability.

The loess hilly area is one of the areas with a high incidence of geological disasters, and it is urgent to use appropriate evaluation factors and training models to conduct research on the susceptibility assessment of geological disasters. Kangdian Town, Gongyi City, the township hardest hit during the “7·20” extremely heavy rainstorm in Zhengzhou, was taken as the study area. Based on satellite remote sensing interpretation, field survey, UAV aerial photography and relevant data collection, an evaluation system covering 13 influencing factors of three main control factors, namely loess interface, human engineering activities and hydrodynamic effects, was constructed. CatBoost model, XGBoost model and LightGBM model were used to carry out the evaluation study of geological disaster vulnerability. Based on the machine learning model with the best performance, SHAP(shapley additive explanations) algorithm was used to complete the global interpretation of characteristics and dependency analysis. The results show that the CatBoost model has higher accuracy than other models (XGBoost and LightGBM), and performs the best in AUC(area under curve) value, accuracy, precision, recall, F₁ score, and field validation. The proportion of areas with extremely high, high, medium, low, and extremely low susceptibility is 3.19%, 1.40%, 2.04%, 5.93%, and 87.44%, respectively. The extremely high and high susceptibility areas are mainly distributed on both sides of gullies with strong human activities, and slope cutting and building are important causes of geological disasters. The aim of this study is to optimize the modeling approach, investigate the uncertainty and interpretability of the modeling process, explain and analyze the decision-making mechanism of machine learning susceptibility, and provide scientific basis for geological disaster prevention and control in the loess hilly area of western Henan.

Natural cooling is one of the more energy-efficient and widely used cooling methods in data center air conditioning systems. Based on the climatic characteristics of Suzhou, Beijing, Guiyang, Guangzhou and Urumqi, the natural cooling system model of data centers was established by TRNSYS, and then the energy saving effect of data centers under the same chilled water supply and return temperatures in different regions was studied. The results show that when the chilled water supply/return temperatures are 15/22 ℃, the longest time for complete natural cooling is in Urumqi, accounting for 67.4% of the year, and the shortest time is in Guangzhou, accounting for 10.2% of the year; the longest time for complete mechanical cooling is in Guangzhou, accounting for 63.0% of the year, and the shortest time is in Urumqi, accounting for 4.3% of the year; the longest time for part of natural cooling is in Guiyang, accounting for 31.8% of the year, and the shortest time is in Beijing, accounting for 4.3% of the year. The longest part of natural cooling time is in Guiyang, accounting for 31.8% of the year, and the shortest is in Beijing, accounting for 18.6% of the year. At full load, the lowest annual average system power usage effectiveness (PUE) was in Urumqi at 1.227, and the highest PUE was in Guangzhou at 1.299. The average annual PUE of the five cities decreases as the load factor increases, and the effect of air conditioning load on the average annual PUE becomes smaller as the load factor increases. The findings provide theoretical support for guiding the application of natural cooling technology in data centers in different regions.

Rock mass classification is a fundamental component in tunnel engineering construction. With the rapid advancement of mechanized and intelligent construction technologies in China, drilling-parameter-based intelligent rock mass classification methods have become crucial in facilitating smart mechanized tunneling. This need is especially pronounced in the mountainous regions of Western China, where complex terrain and challenging construction, combined with limited experience in mechanized tunneling and the restricted applicability of current intelligent rock mass classification methods, make mechanized construction crucial for improving project quality and effectively controlling construction risks. A predictive method was proposed for intelligent rock mass classification using drilling measurement parameters. Focusing on multiple long tunnels as research subjects, on-site drilling parameters were collected and rock mass mechanical tests was conducted to construct a drilling parameter database, then intelligent algorithms was applied, such as support vector regression (SVR) and particle swarm optimization-back propagation (PSO-BP), to develop a predictive model for rock mass classification. The result indicates that the absolute value of correlation coefficient |r_s| between drilling parameters and rock mass classification indices is greater than 0.6, demonstrating a significant correlation, where torque and rotational speed show the strongest correlation with rock mass classification indices. A standardized parameter index database with 574 ideal samples was established through data-cleaning tools. Comparative analysis of predictive accuracy across intelligent algorithms indicated that the PSO-BP model demonstrated the best performance. The PSO-BP neural network-based prediction model was validated by transient electromagnetic (TEM) and tunnel seismic prediction (TSP) advanced geological forecasting, confirming its accuracy in predicting rock mass classification and providing reliable support for mechanized tunnel excavation.

A new anomaly sound detection algorithm was studied that combines attention mechanisms and domain generalization techniques to more accurately identify normal and abnormal sounds in mechanical equipment. Specifically, two neural networks were jointly trained using a sub-cluster Adacos loss function, with features modeled by a Gaussian mixture model (GMM). Anomaly scores were calculated using negative log-likelihood values, and a 90th percentile threshold was set for detection. The algorithm demonstrated strong performance across seven types of machines, including fans and bearings, achieving harmonic mean AUC(area under curve) and pAUC values of 76.69% and 87.99%, with the highest performance observed on valve data. Compared to two baseline systems, the algorithm improved AUC and pAUC by 24.08% and 20.68%, respectively. Ablation studies further confirmed the positive impact of the GMM, attention mechanism, and Scadacos loss function. When tested against eight other algorithms on the same dataset, the proposed method showed a 4.16% improvement in the harmonic mean of AUC and pAUC, highlighting its significant advantage in anomaly sound detection tasks.

In order to improve the accuracy of network traffic classification, a traffic classification method combining an attention mechanism and a convolutional neural network was proposed. An attention mechanism layer was designed and implemented on the basis of the convolutional neural network model, which received the output of the fully connected layer as input, calculated the weight of the input features, and multiplied it by the original features to strengthen the key features. This, in turn, helped to improve the model's ability to capture key information. Secondly, in order to solve the problem that the model was overfitting to the high-proportion category due to the unbalanced sample number of network traffic categories, and it was difficult to identify the small-proportion categories, a method to augment the dataset was proposed. Considering the perspective of hyperparameter combination optimization, a hyperparameter search strategy based on Bayesian optimization and five-fold cross-validation was proposed to optimize the hyperparameter combination of the model. The combination of hyperparameters of the model was determined by the above methods. The public dataset was used for the above experiments and model tests. The results show that compared with other methods, the overall accuracy, precision, and F₁ score are significantly improved, which verifies that the proposed method has better classification performance.

A mathematical model was established to address the multi-constraint, large-scale three-dimensional bin packing problem. A hybrid metaheuristic algorithm combining an improved whale algorithm with simulated annealing was proposed. The algorithm discretized the whale algorithm, including individual encoding and updating mechanisms, and utilized simulated annealing to overcome local optima traps. Moreover, a heuristic loading rule was designed for decoding and optimizing the packing solution. The algorithm was evaluated using standard packing instances from Bischoff and Ratcliff's OR-Library, as well as real-world cargo order data, covering a range of cargo types from weakly heterogeneous to strongly heterogeneous. The proposed algorithm achieved a balance between global and local search capabilities, resulting in high packing efficiency for various types of containers. Specifically, the average container filling rate were 92.24% for weakly heterogeneous cargo, 88.78% for strongly heterogeneous cargo, and an overall average of 91.29%. This result provides valuable insights and references for the study of three-dimensional bin packing problems.

The silica with the highest content in sandstone was taken as the main research object. The SiO₂ surfaces underwent hydroxyl (—OH) and methyl (—CH₃) treatments to represent the hydrophilic and hydrophobic walls, respectively. LAMMPS software was utilized to implement molecular dynamics simulations and replicate the process of supercritical CO₂ extraction of crude oil components. The results indicate that the temperature is 333.15 K and the pressure is 20 MPa. The hydroxylated silica surface extracted 5.07% more saturated hydrocarbon molecules using supercritical CO₂ compared to the methylated silica surface. The interaction energy between the oil components and the two wall surfaces is mutually attractive. The interaction energy of saturated hydrocarbon molecules decreases by 77.52% for hydroxylated silica surface and 46.04% for methylated silica surface, respectively. Additionally, the interaction energy of methylated silica on saturated hydrocarbon is greater than that of the hydroxylated silica surface.It is important to note that carbon dioxide easily extracts saturated hydrocarbons with short molecular chains. The diffusion coefficients of crude oil components under two surface conditions are saturated hydrocarbons > resins > aromatic hydrocarbons > asphaltenes.

In order to strengthen the heat-mass transfer performance of humid air-spray water outside staggered tube bundles (STB), an analytical model was constructed for heat-mass transfer performance of humid air-spray evaporative cooling in staggered tube bundles based on the coupled method of DPM(discrete phase model) and Wall film model. The verification results show that the error was less than 1.1% for simulation and parameter results. Meanwhile, the influences were studied for three key structural parameters on heat-mass transfer performance. The results show that the heat transfer performance is improved between tube wall and spray water with the increase of longitudinal and transverse spacing of tube bundles. However, the mass transfer performance decreases of humid air-spray water with the increase of transverse spacing. Meanwhile, Nusselt number increases by 33.3% with the increase of longitudinal spacing from 30 to 70 mm, increases by 73.5% with the increase of transverse spacing from 10 to 50 mm. Besides, the heat transfer performance proves to be better when contact area increases between tube bundle and spray water with larger pipe diameter. At a certain transverse and longitudinal spacing, the lowest humid air temperature and highest enthalpy are located on the maximum pipe diameter (24 mm), and its decreasing and increasing degrees are 11.2% and 35.6%, respectively. The above results can provide a theoretical basis for optimizing the structure of staggered tube bundles and improving the heat-mass transfer efficiency.

In order to meet the transfer and transportation requirements of passengers and to significantly mitigate the loss of passenger flow that exceeds the waiting tolerance threshold, the optimization method of feeder bus scheduling considering the arrival time and passenger flow loss of rail transit trains was studied. The distribution of passenger flow demand was characterized by the passenger transfer demand and the arrival times of rail transit within the study period. The transfer time was described by the alignment between the time passengers arrive at the station of the bus and the bus departure schedule, as well as the operational capacity of the buses. The constraints of bus departure intervals, passenger flow loss and transfer demands were considered, and the multi-objective optimization with the minimum passenger flow loss, bus number and passenger transfer waiting time was realized under the limited number of buses that can be scheduled. Due to the contradictions among the optimization objectives, the model was solved with Non-dominated Sorting Genetic Algorithm II (NSGA-II). Finally, taking the actual bus routes as an example, the results show that the optimized model takes into account the bus operation cost and the passenger transfer time cost, and can obtain the scheduling that meets the passenger flow demand and represents different priorities. When the number of feeder buses is the same, the total transfer waiting time of the optimized method is reduced by 8.0% compared with the uniform headway. The average factor under the uneven is 59.3%, which is better than the average factor of 50.2% under the uniform. The calculation results validate the effectiveness and rationality of the model and algorithm, effectively enhancing the match between the time and capacity of buses and urban rail transit.