China’s tourism industry has become a strategic pillar industry in China, playing an important role in developing the economy and providing employment. Therefore, how can we avoid or reduce the hazards of tourism emergencies and give full play to the development advantages that tourism brings to the city has become the focus of this paper. In this paper, the objective function is used to construct a two-stage stochastic optimization model without opportunity constraints to minimize the partial cost of the first stage and the expected total cost of the second stage. Considering the problem of maximizing the utilization rate of emergency shelters in tourist attractions, the opportunity constraint model is introduced to help decision makers allocate resources reasonably. Based on the center siting cost and vehicle distribution cost, a mixed integer nonlinear objective function model is constructed and the model is solved using the improved ant colony algorithm. Seven emergency management simulation scenarios are set up to analyze the effect of emergency management by combining simulation and empirical research. The experimental results show that among the emergencies at all levels of the sites in Y scenic area in the past 5 years, the number of level 2 emergencies is the highest, and the average number of emergencies occurred in each site in the past 5 years is 7.48. According to the model’s solution of the site selection results, the emergency center A covers 5 distribution warehouses, and the emergency center B covers 10 distribution warehouses.

The numerical simulation of the velocity decay characteristics of multilayer spherical fragments under bombardment loading is carried out by using LS-DYNA, and the distribution law of the velocity decay characteristics of multilayer spherical fragments is obtained. The ballistic limit (V50) of the multilayer spherical fragment on a 4mm 2024 aluminum target at 90° angle of attack is also obtained by ballistic test. Based on the consistency between the numerical simulation and the test results, the influence of the quality of the multilayer spherical fragment on V50 is analyzed. The air resistance coefficient is calculated with the numerical simulation results by constructing a rag flight distance calculation model. The maximum error between the calculated results and the test results is about 2%, and the theoretical calculated values are in good agreement with the numerical simulation and test results. Under the condition of the same initial velocity, the attenuation coefficient of the spherical fragment in long-distance flight is constant. The aerodynamic drag coefficient is related to the initial velocity of the fragment, which is linearly related to the initial velocity in the range of the design concern of the combat unit (1.2-2.2km/s).

Compared with traditional target detection algorithms, deep learning-based target detection algorithms trained on rich sample data do not need to design features artificially, are better adapted to environmental changes, and the accuracy and efficiency of detection are dramatically improved. This paper relies on the deep convolutional neural network structure to construct the YOLOv5 target detection model. On the input side of the model, three data enhancement techniques, namely mosaic data enhancement, adaptive anchor frame and adaptive image scaling, are adopted respectively to improve the accuracy, generalization ability and detection speed of the model in the target detection task. Attention mechanism is introduced and YOLOv5 framework is improved to construct a new network model. For the efficiency of the target detection task, a loss function is added and a global average pooling operation is used for feature mapping to realize a fully convolutional network. Two widely used evaluation metrics are chosen to evaluate the target detection efficiency of the model. The experiments show that the MAP value of the improved YOLOv5n network model is 2.9979 percentage points higher than that of the original YOLOv5 model, and at the same time, the FPS is substantially improved by 31%. The time taken to complete 100 rounds of training is 20 min, which is 10 min shorter than the pre-improvement algorithm.

The study constructs a solar cell simulation model and tracks the maximum power output from the solar cell using the MPPT algorithm. Simulation simulation experiments are conducted to analyze the effects of changes in environmental factors such as season, weather, light, temperature, wind speed, etc. on the current and power output of solar cells. The total output power and the peak output power of the solar cell are the largest in summer, which are 7407.69kW and 114.93kW, respectively, and the total output power and the peak output power of the solar cell are the smallest in fall, which are 1748.96kW and 31.58kW, respectively. The peak power output of the solar cell is the largest in sunny days, which is 107.56kW, and the smallest in rainy days, which is 37.06kW. The total solar cell power output is maximum (7896.93kW) on clear to cloudy days and minimum (1955.27kW) on rainy days. The solar cell output current and maximum power values decreased with decreasing light intensity. The ambient temperature has little effect on the short circuit current, the output current increases slightly with increasing temperature, the open circuit voltage decreases drastically with increasing temperature and the maximum output power decreases with increasing temperature. The maximum output power of the solar cell increases with increasing wind speed.

MOOC as a new teaching mode is developing in full swing, however, MOOC courses face the thorny problems of high dropout rate and low completion rate. Therefore, this paper selects 12 learning behaviors and uses logistic regression model, decision tree and other methods to predict the withdrawal behavior according to the MOOC data on 365 University platform. The logistic regression prediction is analyzed for prediction accuracy, and its AUC value is 0.83 and 0.75, which proves that the logistic regression analysis can achieve the prediction of MOOC withdrawal behavior more stably and accurately, and helps to provide scientific guidelines for improving MOOC learning mode and learning efficiency. From the case study, it is obtained that among all the learning behaviors, the weight of online rate is 0.7582, which has the highest weight, indicating that the online rate of college students is an important index for judging whether they will produce withdrawal behaviors, which deserves the attention of MOOC platforms and educators.

By optimizing the automation configuration of medium-voltage distribution lines, capturing the initial signals of cable insulation hidden danger, combining the real case data of 6 years of distribution network insulation faults and hidden danger in a city of Zhejiang, summarizing the waveform law and progressive signal characteristics in the process of insulation hidden danger deterioration, a set of real-time monitoring method based on the analysis of big data of the medium-voltage distribution line cable insulation deterioration of the corona hidden danger has been developed. The method is based on the master station to realize localization, instead of periodic on-site equipment charged detection, has been verified on-site and found discharge traces cable head in advance. This method utilizes distribution automation and dispatch automation configurations to capture the instantaneous zero-sequence overcurrent signals corresponding to insulation degradation discharges, waveform characteristics, acoustic mutations, and environmental information as input. A quantitative risk algorithm consisting of eight analysis dimensions such as zero-sequence spike characteristics, number of spikes, and synchronization of acoustic ripple and spike timing is used. Three optional computational media, including master station, enhanced DTU, and DTU external component, are used to give hidden risk localization. The two methods, local discharge detection robot and manual detection, are used to confirm the site and then carry out out outage maintenance to prevent the further expansion of hidden dangers. The method relies on the distribution automation of existing protection devices and master station configuration to assist a small number of sensors and edge computing devices to realize, through the protection device uninterrupted monitoring instead of manual periodic local discharge detection. It solves the problems of high cost of periodic testing, unavoidable accidents caused by continuous insulation degradation in the interval of testing cycle, hidden location of some cables and blind area of testing, and effectively improves the reliability of power supply.

Deep learning-based methods can be combined with skeleton data, but they only consider the feature vectors formed by joint coordinates and do not extract the spatio-temporal dependencies between skeletons. In order to provide a more comprehensive detection and recognition of spatio-temporal relationships in human action sequences, this paper proposes a graph neural network-based human action detection and recognition method by combining YOLOv5, AlphaPose, and spatio-temporal graph convolutional network (ST-GCN) algorithms under the interpretable artificial intelligence (XAI) perspective. Firstly, the improved YOLOv5s target detection algorithm is used to get the human body detection frame and obtain the human body position information, then the AlphaPose pose estimation algorithm is used to obtain the coordinate information of the joint points of the human skeleton, and finally the improved ST-GCN algorithm is used to construct the spatio-temporal graph and extract the spatio-temporal dependencies between the joints to complete the human body action recognition. Through experimental verification, the method can accurately recognize human fall, running, kicking, and squatting actions on the dataset, with a recognition accuracy of 92.04%, and compared with the five baseline models, the method has higher recognition accuracy, with the values of each index greater than 91%, which can provide technical support for human behavior recognition.

The development of society and the change of the times have brought some degree of change to the development of preschool music classroom. This paper introduces the OBE concept into the education of preschool music course, designs the teaching objectives of the music course according to the guidance of the concept in order to realize systematic teaching, and analyzes the students’ cognition of various dimensions in the preschool music course by using the cognitive level diagnosis method. Based on this method and the Hadoop system, a big data platform for integrated teaching of preschool music course is constructed, and teachers are assisted to intervene in teaching through the platform’s teaching data query, statistics and analysis functions, so as to realize the integrated teaching mode of preschool music course and mathematical statistical analysis. The results of the teaching practice show that after the implementation of the integrated teaching mode, students’ learning attitude towards the preschool music course and their knowledge of music-related knowledge increased significantly (P<0.05), and the level of independent and inquiry learning was also improved. This study can make the teaching of preschool music course more meaningful, more adaptable to the needs of talent training in today's society, and create an integrated teaching curriculum that is more conducive to the cultivation of students' musical literacy and interest.

ETC gantry data and other monitoring data provide data support for highway traffic flow prediction, for this reason, this paper proposes an attention mechanism-driven traffic flow prediction model to scientifically coordinate and schedule highway traffic conditions. Based on the fusion of multivariate monitoring data, the model utilizes ConvLSTM to generate global location coding, learns the data characteristics through the jump expansion attention structure, and completes the traffic flow prediction using the mask attention structure. The example analysis verifies that the predicted values of traffic flow and speed of this paper’s model are closer to the real values, and compared with the models such as ARIMA, LSTM and BiLSTM, this paper’s model has lower values of RMSE and MAE indexes in the prediction of traffic flow and speed, and the prediction error is smaller. The article also validates the model’s prediction under 5min, 15min and 30min prediction lengths, showing that the model has excellent performance and good prediction stability.

With the continuous development of high-power laser equipment and the continuous expansion of the scope of the application platform, the demand and application of high-power laser equipment in various fields are becoming more and more extensive, and its output power has also put forward higher requirements. In order to promote the development of high power laser equipment toward higher energy conversion efficiency, research and design temperature control device to manage the waste heat generated in the energy conversion process of high power laser equipment. On the basis of PID control algorithm using LADRC algorithm, rapid realization of temperature precision control, so as to enhance the energy conversion efficiency of high-power laser equipment. When the temperature control device in the temperature control range of 10 ℃ ~ 40 ℃, the temperature control accuracy is better than ± 0.03 ℃, and in 144s to reach the set temperature, the temperature control overshoot is less than 2.33%, to meet the requirements of the laser working temperature control in the working process of high-power laser equipment, and to lay the foundation for the realization of high energy conversion efficiency. Compared with the modified PID controller, the energy conversion efficiency is relatively improved by 1.57%. The temperature control device designed based on the improved PID control algorithm in this paper can significantly improve the energy conversion efficiency.