With the development of big data and education informatization, education reform and talent cultivation mode are facing digital reform. In this paper, the important feature selection algorithm based on random forest is used to select the relevant features that affect the application effect of teachers’ practice teaching cultivation and innovation mechanism, which lays the foundation for constructing the practice teaching data mining model based on Light GBM. Then the data processed by feature selection is preprocessed and standardized, and then the processed data is partitioned and the model is trained in turn to get the prediction results. The Light GBM-based practical teaching data mining model was compared with other classification models in different datasets, and the experimental results showed that the model in this paper has an advantage over other classification models in a number of evaluation indexes, with the highest accuracy rate of 13.07%, and the model data mining results accurately locate the open innovation experimental indexes that have a lower score of importance to students’ development, and provide a good basis for the optimization of teaching paths and students’ development. , which provides ideas for the optimization of teaching paths and the improvement of the impact of students’ future development.

Sparse decomposition has been generally emphasized in signal processing theory. In this paper, a nonelectrical signal feature dataset of key components of high-voltage DC converter valve is established by using principal component analysis to streamline the data volume. The compression-aware feature extraction algorithm based on polynomial matrix sparse coding is used to extract and collect the nonelectrical signal parametric data. Through the performance over the experimental signal analysis, it can be known that the eigenvalues of a total of 10 parameters, including the infrared temperature measurement results, the appearance, the presence of corrosion or dirt, and the presence of abnormal vibration and sound, are all greater than 1. Therefore, these 10 parameters are identified as the key parameters. When the number of measurement points is between 64 and 200, the algorithm in this paper can satisfy the need of feature extraction when the signal length is insufficient, compared with the traditional approach. In the empirical analysis of the vibration signal as an example, the method of this paper can effectively extract the frequency and time domain of the vibration signal.

As the core equipment of high-voltage direct current transmission system, the operation status of the converter valve directly affects the safety of the power grid. In this paper, we first construct a multisource data fusion system to realize the error-free fusion of fault information parameters. Then, combined with the random forest algorithm, the time-varying law of the electrical characteristics of the converter valve based on harmonic theory is extracted. Finally, the collected time-varying laws of electrical characteristics are input into the constructed Random Forest particle swarm optimization model, and the trained model is used to monitor the status of the converter valve. In the simulation experiment, the ±800kV UHV DC transmission system is built by PSCAD/EMTDC software, from which the current waveforms are collected when the converter valve fails, the time domain features of the current are extracted, and the obtained converter feature indicators are selected using the Random Forest algorithm, and 10 important features will be finally identified to construct the converter valve feature indicator set, and input into the Random Forest Particle Swarm Optimization model and the other comparative models for training and testing. The accuracy of this model is 97.5%, which is better than other comparative models. The study provides a high-precision solution for converter valve condition monitoring and effectively extends the application of multi-source data fusion in power equipment.

This paper proposes a real-time computational method for multidimensional dynamic data fusion (VIO-SLAM) for intelligent monitoring of seat belts in the grid construction environment. In this paper, the optical flow method is first used to process and track point features, and the geometrically constrained line matching algorithm is utilized to improve the accuracy of feature matching. Combined with IMU modeling and pre-integration techniques, it effectively reduces the computation of high frequency IMU data and improves the system efficiency. At the same time, a real-time lightweight semantic segmentation system is constructed to achieve fast semantic understanding of the construction scene. The real-time and accuracy of data processing is further improved by sliding window method with BA optimization. On this basis, a VIO-SLAM algorithm based on EKF fusion of multidimensional dynamic data is proposed to realize real-time monitoring and localization of seat belt status. The results show that when a dangerous collision occurs in a complex power grid construction environment, the protection performance of shoulder belt, neck bending moment force and head acceleration of the construction personnel under the method of this paper is much higher than that of the traditional seat belt. In the process of emergency collision avoidance, the VIO-SLAM algorithm is able to tighten the seat belt in advance for the construction personnel, which has better protection performance and can achieve the purpose of “collision avoidance and damage reduction”. The pre-tensioning force for eliminating the gap in the webbing of seat belts and the pre-tensioning force for somatosensory warning reminders are also determined to improve the protection performance of construction workers.

At present, the physical training of public security police has not formed a unified training system in the country, and various places ignore the cultivation of other aspects of the ability to take skill training as the leading role, and solve the problem of how to train through the construction of the system, so as to ensure that the physical training of public security police is carried out effectively. This paper explores the impact of physical training on college students’ professionalism in public security colleges, constructs the K nearest neighbor classification algorithm, and introduces the relevant activation function to deal with more complex students’ physical training exercise trajectories. ATT-DAN multitarget tracking model is constructed to extract the feature information of college students’ physical fitness training, obtain the target movement trajectory, and parameterize the representation of students’ physical fitness training programs. The correlation ranges of frequency, average score, highest grade score of physical fitness training and occupational ability were between 0.415~0.632, 0.452~0.769, 0.412~0.715, respectively, and the credibility and stability of the occupational ability characteristics were good. Meanwhile, the linear regression of the two showed that the correlation P value of age, 30-second deep squat, pull-up, 3200 meters, and 15-second repetitive straddle with occupational ability was less than 0.05, and there was a positive correlation between the two.

Power system simulation training is one of the important means to improve the quality of operators and ensure the safe and effective operation of power systems. Research based on digital twin technology, combined with configuration algorithms to give the substation integration diagram model generation method, developed a smart substation virtual training system. The intelligent monitoring is studied, the digital twin-based substation output voltage anomaly detection method is designed using the tracking differentiator method, and finally the simulation test of the intelligent substation virtual training system is carried out. The analysis shows that the voltage anomaly detection method in this paper is highly accurate and can extract the voltage anomaly waveform, and the offset rate of its collected signal is significantly lower than that of the comparison method (11.58%~14.84%), which is only 0.54%. The training test of fast distance protection, differential protection and zero sequence protection verifies the feasibility and effectiveness of the virtual training system in practical application. The platform can effectively promote the reform of applied electric power practice courses and provide a backbone for the training of new power system talents.

The rapid development of information technology has put forward higher requirements for teachers, and the traditional training model is difficult to meet the demand. The article constructs a teacher digital competency framework based on the ASTD model, realizes the division of teachers’ professional competence, and explains the professional core connotation of teacher digital competency in detail. A personalized resource recommendation model for teachers is constructed using artificial intelligence technology, which provides accurate recommendations for teachers through candidate resource extraction and learning resource screening. At the same time, with the help of Google Cloud Services digital tools, the design of teachers’ digital teaching and research activities was accomplished, and communication and cooperation with users in the virtual community was promoted. The combination of the two is integrated into the development of teachers’ professional skills to enhance their teaching competence. The mean values of accuracy, applicability, timeliness, personalization, and diversity of learning resource recommendations under artificial intelligence technology ranged from 4.123 to 4.544, with good recommendation performance. The Google Cloud Services platform can promote teaching and research exchange activities among teachers. The use of artificial intelligence and digital tools makes teachers improve their professional skills in knowledge base, instructional design, teaching and research between 24.04% and 91.00%, and with their intervention, teacher competency shows significant improvement.

Under the environment of plateau alpine region, the new model of substitute construction separating government construction and management functions has gained great development in barracks construction, which significantly improves the risk management level of barracks facilities to some extent. From the significance of barracks facilities construction guarantee in highland alpine area, the article proposes a risk identification framework for the substitute construction unit of Someplace facilities in highland alpine area based on the whole life cycle of engineering projects. Combined with the risk identification framework, the risk evaluation index system of the agency construction unit is constructed, and then the AHP hierarchical analysis method is introduced to solve the weight of the indexes, and combined with the fuzzy comprehensive evaluation method, the AHP-FCM evaluation model is constructed. A barracks facilities project in a camp area is selected as a case study, and Company T is used as the research object to carry out data analysis of its risk degree using the AHPFCM model. In the construction of barracks facilities in highland and alpine areas, the biggest risk faced by the construction unit is the project implementation stage, the weight of which reaches 29.93%, and the fuzzy comprehensive evaluation of Company T’s risk score is 3.182, which is between medium and large risks. Therefore, the agency needs to examine and check its own risk factors in time, in order to lay a solid foundation for ensuring the smooth implementation of the agency project of barracks facilities in highland alpine areas.

The study is based on the important role of graph theory in the teaching of physical dance and aesthetic education, integrating the concept of graph theory into it and designing the training path of physical dance and aesthetic education based on graph theory. Taking two classes in a university as the research object, the teaching experiment is conducted to compare their physical quality and course performance after the experiment, and the aesthetic education evaluation index system is constructed, and the index weights are measured using the combination assignment method to carry out the comprehensive scoring. After the experiment, the students improved in physical quality, course grades and aesthetic effect, and as far as the students of traditional teaching class are concerned, the experimental students improved in course grades and aesthetic effect by 18.17% and 7.52% respectively. The teaching practice of integrating the concept of graph theory and the curriculum of physical education dance and aesthetic education not only embodies the concept of cross-disciplinary teaching, but more importantly improves the physical quality, physical education dance level and aesthetic effect of students in colleges and universities, and provides a reference for the teaching reform of physical education dance and aesthetic education in colleges and universities.

Transportation demand is gradually increasing and road traffic congestion is becoming more and more serious. Traffic state prediction is one of the important bases for accurate traffic management and control. This paper investigates a traffic state prediction method based on a deep learning algorithm fusing spatio-temporal graphical convolutional networks, and explores the law of path selection decision-making of pedestrians under different traffic flow prediction and guidance strategies, and analyzes the effect of the implementation of the information guidance policy by traffic managers in realistic scenarios using evolutionary game theory. The simulation results combined with the traffic simulation model show that the traffic state prediction method proposed in this paper is more effective compared with other models. The evolution results are more reasonable when the value of the path adjustment rate in the replicated dynamic model is the inverse of the number of iterations. In the perceptual error analysis, when the value of perceptual error 1 is taken to be too large, i.e., when the perceptual error of the first type of travelers is small and small, it tends to be a deterministic choice. Finally, a traffic simulation model is implemented to validate the performance of the proposed model and propose congestion mitigation strategies.